A method and system for automatically constructing a teaching character motion effect pairing dataset

Abstract

This invention relates to the field of data processing, and more particularly to an automated method and system for constructing a teaching text-motion effect pairing dataset. This application acquires raw teaching text data and preprocesses it to obtain multiple text units; then, it performs semantic parsing on each text unit to identify motion effect trigger information and generate a set of motion effect semantic tags; next, it calls a pre-built motion effect parameterization template library to match initial motion effect parameters; then, it analyzes the association features of the text units, adjusts the initial motion effect parameters, and generates a motion effect configuration file; subsequently, it renders the motion effect configuration file to generate a dynamic video, and performs consistency verification with the corresponding text units. The text units, motion effect configuration files, and dynamic videos that pass the verification are associated and stored as paired samples; finally, the paired samples are labeled and data augmented to generate a teaching text-motion effect pairing dataset. This invention achieves automated construction of teaching text-motion effect pairings, improving the efficiency and quality of dataset construction.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to an automated method and system for constructing a teaching text animation matching dataset. Background Technology

[0002] In educational scenarios such as teaching materials and online courses, adding appropriate animation effects to text content can improve learning attention and comprehension efficiency. Current technologies primarily rely on manual or semi-automated methods to construct paired datasets of teaching text and animation effects, lacking a complete automated process from text semantic parsing, animation parameter matching and adjustment, rendering verification to data augmentation. Existing methods often treat text analysis and animation generation separately, resulting in low dataset construction efficiency, poor scalability, and a lack of structured animation parameter descriptions in the generated paired samples, making it difficult to support the training needs of downstream teaching animation recommendation or generation models.

[0003] Secondly, teaching texts often contain complex sentences with multiple clauses, each carrying different animation intentions. Existing technologies typically treat the entire sentence as a whole, outputting only a single animation tag, leading to the neglect of some animation intentions within complex sentences. Furthermore, existing methods lack mechanisms for expressing the temporal relationships between animations in clauses, such as sequential or simultaneous playback. Additionally, existing technologies typically calculate animation duration linearly based on text length. However, when the text is extremely short, the linearly calculated duration may fall below the rendering limit of the animation template; when the text is extremely long, the calculated duration may exceed the reasonable range for the animation type. Existing methods struggle to handle duration in these extreme cases, potentially resulting in some generated animation pairs being visually unusable. Therefore, existing technologies have shortcomings and require improvement. Summary of the Invention

[0004] To address one or more problems in the existing technology, the main objective of this application is to provide an automated method and system for constructing a teaching text animation matching dataset.

[0005] To achieve the aforementioned objectives, this application proposes an automated method for constructing a teaching text animation matching dataset, the method comprising: Obtain the original teaching text data, and preprocess the original teaching text data to obtain multiple text units; Semantic parsing is performed on each text unit to identify motion effect triggering information in the text unit. Based on the identification results, a corresponding set of motion effect semantic tags is generated. The motion effect triggering information is a text feature that can be mapped to motion effect type, motion effect intensity, or motion effect target. Call the pre-built motion effect parameterization template library and match the motion effect semantic tag set with the corresponding initial motion effect parameters from the motion effect parameterization template library; Analyze the association features of the text unit, adjust the initial animation parameters according to the association features of the text unit, and generate an animation configuration file corresponding to the text unit; The motion effect configuration file is rendered to generate a dynamic video, and the consistency between the dynamic video and the corresponding text unit is checked. The text unit, motion effect configuration file and dynamic video that pass the check are associated and stored as a paired sample. The paired samples are labeled and augmented to generate a teaching text animation pairing dataset.

[0006] This application also provides an automated system for constructing a teaching text animation matching dataset, including: The acquisition module is used to acquire raw teaching text data and preprocess the raw teaching text data to obtain multiple text units. The parsing and recognition module is used to perform semantic parsing on each text unit, identify the motion effect triggering information in the text unit, and generate a corresponding motion effect semantic tag set based on the recognition result. The motion effect triggering information is a text feature that can be mapped to the motion effect type, motion effect intensity, or motion effect target. The calling module is used to call the pre-built motion effect parameterization template library and match the motion effect semantic tag set with the corresponding initial motion effect parameters from the motion effect parameterization template library; The analysis and generation module is used to analyze the association features of the text unit, adjust the initial animation parameters according to the association features of the text unit, and generate an animation configuration file corresponding to the text unit. The storage module is used to render the motion effect configuration file to generate a dynamic video, and to perform consistency verification between the dynamic video and the corresponding text unit. The text unit, motion effect configuration file and dynamic video that pass the verification are associated and stored as a paired sample. The enhancement processing module is used to label and augment the paired samples to generate a teaching text animation pairing dataset.

[0007] This application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods described above.

[0008] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0009] This application presents an automated method and system for constructing a teaching text animation matching dataset. The method establishes a fully automated process from raw teaching text to animation matching data. First, through preprocessing and semantic parsing, the implicit animation intent within the teaching text is identified, eliminating the need for manual judgment of animation types and improving dataset construction efficiency. Second, by calling a parameterized template library for animation matching and utilizing the association features of text units to adjust the initial parameters contextually, the duration of the generated animation matches the reading rhythm of the text, while avoiding visual conflicts between adjacent animations, thus ensuring the display quality of the paired samples. Third, post-rendering consistency verification automatically filters out inferior samples with semantic mismatches or abnormal parameters, ensuring that every pairing in the dataset has practical usability. Finally, the annotation and data augmentation steps not only add structured metadata to the samples but also expand the sample size through text rewriting and parameter perturbation, enhancing the dataset's diversity and generalization ability. In summary, this method enables the large-scale, standardized, and automated construction of a teaching text animation matching dataset, providing high-quality basic resources for animation recommendation or generation model training in subsequent intelligent education scenarios, reducing manual costs and enhancing the practical value of the dataset. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating an embodiment of the automated construction method for a teaching text animation matching dataset according to this application. Figure 2 This is a flowchart illustrating an embodiment of the automated construction method for a teaching text animation matching dataset according to this application. Figure 3 This is a schematic block diagram of the structure of an automated construction system for a teaching text animation matching dataset according to an embodiment of this application; Figure 4 This is a schematic block diagram of the structure of a computer device according to an embodiment of this application; Figure 5 This is a schematic diagram showing the misalignment between the animation duration and reading time before and after text rewriting enhancement according to an embodiment of this application.

[0011] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0013] Reference Figure 1This application provides an automated method for constructing a teaching text animation matching dataset, the method comprising: S1. Obtain the original teaching text data, and preprocess the original teaching text data to obtain multiple text units; S2. Perform semantic parsing on each text unit, identify the motion effect triggering information in the text unit, and generate a corresponding motion effect semantic tag set based on the identification result. The motion effect triggering information is a text feature that can be mapped to the motion effect type, motion effect intensity, or motion effect target. S3. Call the pre-built motion effect parameterization template library and match the corresponding initial motion effect parameters from the motion effect semantic tag set in the motion effect parameterization template library; S4. Analyze the association features of the text unit, and adjust the initial animation parameters according to the association features of the text unit to generate an animation configuration file corresponding to the text unit; S5. Render the motion effect configuration file to generate a dynamic video, and perform consistency verification between the dynamic video and the corresponding text unit. Store the text unit, motion effect configuration file and dynamic video that pass the verification as a paired sample. S6. Label and augment the paired samples to generate a teaching text animation pairing dataset.

[0014] As described in steps S1-S3 above, step one, in constructing the teaching text-motion matching dataset, first requires obtaining the original teaching text data. This text data can come from various sources, such as PPT slides used by teachers, subtitle files from online course platforms, chapter content of electronic textbooks, or manually edited classroom transcripts. The obtained original text usually contains a lot of noise, such as extra spaces, line breaks, abnormal use of punctuation, and possible formula placeholders or image reference marks. This noise will interfere with subsequent semantic analysis, so preprocessing is required. The core objective of preprocessing is to segment the original text into several independent text units. The basic basis for segmentation is punctuation marks such as periods, question marks, and exclamation marks that indicate the end of sentences. Semicolons, colons, and conjunctions indicating transitions or causal relationships can also be used for auxiliary segmentation. After segmentation, each text unit corresponds to a relatively complete and semantically independent short sentence or clause. In addition, the preprocessing stage also cleans the text, such as removing meaningless whitespace characters, converting full-width punctuation to half-width punctuation, and retaining symbols with special meanings in teaching scenarios, such as arrow symbols, superscript and subscript marks in mathematical formulas, and circled numbers. These special symbols may play a key role in subsequent motion effect mapping; for example, arrows themselves can trigger movement-type motion effects. After obtaining the text units in step two, it is necessary to parse out the semantic information related to motion effects. Teaching texts often contain rich motion effect intentions. For example, when words like "attention" or "important" appear in the text, it usually means that emphasis-type motion effects are needed to attract learners' attention; when sequential words such as "first step," "second," and "last" appear in the text, it is often necessary to use sequential motion effects to show the progressive relationship of the steps; when the text ends with a question mark, it may be necessary to flash or enlarge the question mark to stimulate thinking. The specific approach to semantic parsing is to first identify the sentence structure skeleton of the text unit, that is, to determine whether the sentence is a declarative sentence, interrogative sentence, exclamatory sentence, or imperative sentence. Different sentence structures correspond to different animation tendencies. For example, interrogative sentences naturally evoke curiosity and are suitable for flashing or scaling animations. Then, pre-defined animation trigger keywords are extracted from the text. These keywords can be pre-collected in a dictionary, which covers common animation hints used in teaching scenarios. Finally, the sentence structure and keyword information are combined, and a pre-defined mapping rule table is used to determine the appropriate animation type and intensity level. For example, "first step" with a sequential sentence structure can be mapped to a medium-intensity slide-in animation; "pay special attention" with an emphatic sentence structure can be mapped to a high-intensity breathing light flashing animation. After semantic parsing, each text unit generates an animation semantic tag set, which includes the animation type, intensity level, and target object.The target of the action is usually a noun phrase in the text that is emphasized or needs to demonstrate an action, such as "cell" in "cell division" or "cell". Figure 2 The "image" in the text. This tag set is the core basis for subsequent animation parameter matching. Text features refer to the linguistic elements with semantic or structural characteristics that can be extracted from text units, and there are pre-defined correlation rules between them and animation effects. The following will explain these three types of mapping relationships in detail in the context of teaching scenarios. The first type of mapping is text features mapped to animation effect types. Different text features often correspond to different animation effect categories. For example, when the text contains sequential indicator words such as "first step", "second", "last", these features are mapped to sequence slide-in or sequential appearance animation effects. When the text contains emphasis words such as "note", "important", "please remember", these features are mapped to blinking, breathing light, or scaling animation effects. When the text ends with a question mark or contains interrogative words such as "why" or "how", the sentence structure features are mapped to strong interrogative effects. The first type of mapping is mapping text features to animation intensity. For example, a blinking or enlarging question mark. When the text contains arrow symbols, the symbol's features are mapped to directional or path-moving animations. The second type of mapping is mapping text features to animation intensity. Some text features not only determine the type of animation but also imply its intensity. For example, "pay special attention" has a higher semantic intensity than "pay attention," and can be mapped to high-intensity blinking; while "pay a little attention" is mapped to low-intensity blinking. The number of exclamation marks or modal adverbs such as "very" or "extremely" can also be used to adjust the intensity level of the animation. Through this mapping, the system can generate animation effects that match the sentiment or emphasis of the text. The third type of mapping is mapping text features to the object of the animation. The object refers to the specific text segment or area in the text that the animation applies to. For example, in "Please observe..." Figure 2 In the phrase "the curve in the middle," the animation trigger information "observation" is mapped to the object being observed. Figure 2The curve in "Note this formula"; in "Note this formula", "Note" is mapped to the object "this formula". The system can extract the object from the noun phrases modified or pointed to by the text features through dependency parsing or pattern matching. Step 3: The motion effect parameterization template library is a resource library pre-established in this application, which stores a variety of reusable motion effect templates. Each template contains all the parameters required to complete a motion effect. For example, a fade-in motion effect requires configuration of fade-in duration, start transparency, end transparency, easing curve type, etc.; a slide-in motion effect requires configuration of slide-in direction, slide-in distance, duration, easing curve, etc.; a blinking motion effect requires configuration of blinking period, blink count, amplitude, etc. These templates adopt a parameterized description method and are usually stored in JSON or XML format for easy reading by the program. And modifications. Once the system obtains the set of semantic tags for a text unit's animation, it matches the animation type and intensity level from the template library. For example, if the tag requires a high-intensity blinking animation, the system will find a blinking template in the library and adjust parameters such as amplitude and blink count according to the intensity level to generate initial animation parameters. If the tag contains information about the target object, this information will also be added to the animation parameters so that subsequent rendering knows which text area the animation should apply to. It's important to note that the templates in the library are independent of specific text; the same template can be reused by different text units, requiring only adjustments to some parameters based on the semantic tags. This design ensures the standardization and consistency of animations while significantly reducing computational overhead.

[0015] As described in steps S4-S6 above, the initial animation parameters obtained in step four are generated independently based on a single text unit, without considering the context of that text unit within the original teaching text. However, in actual teaching demonstrations, the duration and timing of the animation must be coordinated with the reading rhythm of the text and the animations of the preceding and following texts; otherwise, a sense of disjointedness or visual conflict will occur. Therefore, this step requires analyzing the relational characteristics of the text units, which mainly include two aspects. The first aspect is the length characteristic of the text unit itself, usually measured by the number of effective characters. The reading time of a piece of text is roughly proportional to its length, so the duration of the animation should also be adjusted accordingly. For example, a very short sentence may only require a 0.5-second fade-in animation, while a longer paragraph may require 2 seconds or even longer animation duration. The second aspect is the animation information of adjacent text units, that is, what kind of animation has been configured in the previous text unit of the current text unit, and what its start and end times are. This information can be used to adjust the start time and type of the animation of the current text unit. For example, if the animation of the previous text unit has not finished, the animation of the current text unit should start with a delay. Or, if two animation types have visual conflicts, such as simultaneous fade-in and zoom, the system can replace one of the animation types with another non-conflicting type. After the above adjustments, key parameters such as duration, start time, and type in the initial animation parameters are modified to values ​​that adapt to the context. These adjusted parameters, along with other unmodified parameters, are integrated into a complete animation configuration file. The animation configuration file is a structured data object that contains all the information needed by the rendering engine and can be directly handed over to the rendering module for execution. After the animation configuration file is generated in step five, it needs to be rendered into a visual dynamic video. The rendering engine can use a browser-side rendering solution based on HTML5Canvas or WebGL, or it can use cross-platform animation libraries such as Lottie. During rendering, the animation is generated frame by frame according to the parameters in the configuration file and superimposed on the background of the text display. The final output is a short video file, which can be in common formats such as MP4, GIF, or WebM. The video resolution, frame rate, background color, and other rendering parameters can be preset to uniform values ​​to ensure the consistency of the entire dataset. After rendering, a consistency check is required between the generated dynamic video and the original text units. This check aims to filter out samples that contain errors during the pairing process, such as a complete mismatch in animation type due to semantic parsing errors, or animation durations that are too short or too long due to abnormal parameter adjustments. The check can be performed using a combination of methods. For example, it can calculate the ratio of the estimated reading time of the text unit to the actual animation duration; if the ratio exceeds a reasonable range, it is considered unqualified. Alternatively, a pre-trained visual language model can be used to calculate the similarity between the video content and the text semantics; samples with excessively low similarity are also filtered out.Only paired samples that pass the consistency check are retained. Each retained sample contains three components: the original text unit, the motion effect configuration file, and the rendered dynamic video. These three are stored together to form a complete pairing record. In step six, although the previous step yielded quality-screened paired samples, the number of these samples may still be limited, and the descriptive information of each sample is not yet structured enough, which is not conducive to subsequent model training. Therefore, this step performs two aspects of processing: annotation and data augmentation. Annotation refers to adding structured metadata to each paired sample. Specifically, the system extracts the content of the text unit, the motion effect type, motion effect duration, motion effect start time, target, and dynamic video file storage path from the paired sample, and organizes this information according to a preset field structure to form a complete annotation record. This annotation information can be stored separately as a JSONL file, with each line corresponding to the annotation of one sample. Data augmentation refers to generating more variant samples manually or automatically without changing the core semantics of the samples, thereby expanding the scale of the dataset. This application provides two augmentation methods. The first method is text rewriting enhancement, which involves replacing original text units with synonyms or transforming sentence structures to generate new text units. These new text units are then associated with and stored along with the motion effect configuration files and dynamic videos from the original samples, forming a new positive sample. The second method is parameter perturbation enhancement, which involves randomly fine-tuning one or more parameters in the motion effect configuration file. For example, multiplying the duration by a random coefficient between 0.8 and 1.2, or adding a small offset to the movement direction. The perturbed configuration file is then used to re-render and generate a new video, which is then associated with and stored along with the original text units. After annotation and data augmentation, the originally limited number of paired samples is expanded into a large-scale, structured, and diverse teaching text-motion effect pairing dataset. This dataset can be directly used to train motion effect recommendation or generation models, and can also serve as a benchmark resource for the automated production of teaching courseware.

[0016] As described above, this method constructs a fully automated process from raw teaching text to motion effect matching data. First, through preprocessing and semantic parsing, the implicit motion effect intentions within the teaching text are identified, eliminating the need for manual judgment of motion effect types and improving dataset construction efficiency. Second, by calling a parameterized template library for motion effect matching and utilizing the association features of text units to adjust the initial parameters contextually, the duration of the generated motion effects matches the reading rhythm of the text, while avoiding visual conflicts between adjacent motion effects, thus ensuring the display quality of the paired samples. Third, post-rendering consistency verification automatically filters out inferior samples with semantic mismatches or abnormal parameters, ensuring that every pairing in the dataset has practical usability. Finally, the annotation and data augmentation steps not only add structured metadata to the samples but also expand the sample size through text rewriting and parameter perturbation, enhancing the dataset's diversity and generalization ability. In summary, this method achieves large-scale, standardized, and automated construction of teaching text motion effect matching datasets, providing high-quality foundational resources for subsequent motion effect recommendation or generation model training in intelligent education scenarios, reducing manual costs and enhancing the dataset's practical value.

[0017] Reference Figure 2 In one embodiment, the steps of performing semantic parsing on each text unit, identifying motion effect triggering information in the text unit, and generating a corresponding motion effect semantic tag set based on the identification results include: S21. Identify the sentence structure skeleton of the text unit, determine the semantic function type of the text unit based on the identification result, and extract the preset animation trigger keywords from the text unit; S22. Based on the combination of the semantic function type and the motion effect triggering keyword, find the corresponding motion effect type and motion effect intensity from the preset mapping rules; S23. Based on the found motion effect type and motion effect intensity, generate motion effect semantic tags that include the motion effect type, motion effect intensity, and the object to be applied in the text unit.

[0018] As described above, the first step in the semantic parsing sub-step is to understand the sentence structure of the text unit. Sentences in teaching texts are not uniform; different sentence structures carry different expressive intentions. For example, declarative sentences are used to objectively describe facts, interrogative sentences are used to stimulate thinking, exclamatory sentences are used to express strong emotions, and imperative sentences are used to issue commands. These different semantic function types naturally correspond to different animation tendencies. An interrogative sentence might need a flashing question mark to attract students' attention, while an emphatic sentence might need an enlarged action to highlight keywords. The specific approach to identifying the sentence skeleton is to utilize syntactic analysis techniques from natural language processing. The system analyzes the part-of-speech sequence and dependency relationships of the text unit to determine the main structure of the sentence. For example, sentences ending with a period and having a complete subject-verb-object structure are usually identified as declarative sentences, while sentences ending with a question mark or containing interrogative words are identified as interrogative sentences. Simultaneously, the system also extracts pre-set animation trigger keywords from the text. These keywords are pre-compiled into a dictionary covering common animation cues used in teaching scenarios, such as "first step," "attention," "in comparison," "in short," "please look," and "key point." Each word in the dictionary is labeled with its preferred animation type. For example, "Step 1" tends to be a sequential animation, while "Note" tends to be an emphasis animation. This makes implicit information in natural language explicit. A text may appear to be a plain narrative, but the hidden sentence structures and keywords directly reflect the teacher's animation intentions. Through systematic identification and extraction, information that originally required manual judgment is transformed into structured features that can be processed by machines. After extracting the semantic function type and animation trigger keywords in sub-step two, the next step is to combine these two dimensions of information to find a specific animation type and intensity level. Information from a single dimension is often insufficient for accurate judgment. For example, the keyword "Step 1" might match a sliding animation if it appears in a sequential sentence, but if it appears in an interrogative sentence, it might just be a reference rather than a true sequential instruction. Therefore, both need to be used in combination. The preset mapping rule table is a predefined knowledge base containing rule entries that map semantic function type plus keyword combination to animation type plus intensity level. For example, when the semantic function type is a sequential sentence and the keyword is "Step 1," the mapping result is a medium-intensity sequential sliding animation. When the semantic function type is an emphatic sentence and the keyword is "pay special attention," the mapping result is a high-intensity breathing light flashing animation. When the semantic function type is a declarative sentence and there are no obvious keywords, it may be mapped to a default low-intensity fade-in animation. These rules can be obtained through statistical induction from a large number of teaching cases, or they can be manually annotated by educational experts. This solves the core mapping problem from text features to animation parameters. After finding the animation type and intensity in sub-step three, it is also necessary to determine which part of the text this animation specifically acts on. The target is usually the core noun phrases in the text that are emphasized or need to be acted upon. For example, for the text "pay special attention." Figure 2The curve in the image, the object of the motion effect should be... Figure 2 The object of the animation is the noun phrase "curve," not the entire sentence. A common method for determining the object is to perform dependency parsing to identify the noun component modified or pointed to by the keyword. If there is no explicit object or modifier in the text, the entire text unit can be used as the object. The final generated animation semantic tag is a structured data object containing three fields: animation type, intensity level, and object. This tag can be directly used in the animation parameter matching step. For example, a tag might be recorded as animation type "flickering," intensity "high," and object "formula." The subsequent template matching module sees this tag, retrieves the flickering template from the template library, adjusts the amplitude and frequency of the flickering according to the high intensity, and finally applies the animation to the text area "formula."

[0019] In one embodiment, the steps of analyzing the association features of the text unit, adjusting the initial animation parameters based on the association features of the text unit, and generating an animation configuration file corresponding to the text unit include: Obtain the length feature of the text unit, and adjust the animation duration in the initial animation parameters according to the length feature; Obtain the animation information of the adjacent text units of the text unit in the original teaching text, and adjust the animation start time and animation type in the initial animation parameters according to the animation information of the adjacent text units; The adjusted animation duration, animation start time, and animation type are integrated into the animation configuration file.

[0020] As mentioned above, sentences of different lengths in the teaching text of sub-step one require different animation durations. A very short prompt, such as "Please look" or "Pay attention," only needs a quick flash or a brief blink; a longer animation duration would appear cumbersome. A longer definition or theorem statement, such as "The sum of the interior angles of a triangle equals 180 degrees," requires a longer animation display time to allow learners sufficient time to read the text and understand its meaning. Therefore, it is first necessary to obtain the length characteristics of the text unit. The length characteristic can be measured by the number of effective characters, which refers to Chinese characters, numbers, letters, and meaningful special symbols excluding punctuation marks and spaces. After obtaining the length characteristics, the system adjusts the animation duration according to a preset duration calculation formula. The usual practice is to set a unit character reading time, for example, 0.3 seconds per Chinese character, and then multiply the number of effective characters by the unit reading time to obtain the baseline duration. However, when the text is extremely short or extremely long, the result of linear calculation may not be applicable. For extremely short texts, such as a response with only an exclamation mark or a single word, the linearly calculated duration may be lower than the minimum renderable duration of the animation template. In this case, the duration is directly set to the minimum duration supported by the template. For extremely long texts, such as paragraphs exceeding fifty words, the linearly calculated duration may far exceed a reasonable range, resulting in overly slow animations. In this case, the system will limit the duration to a preset maximum value. This boundary handling ensures that regardless of the text length, the generated animation duration is within a visually acceptable range. Through this adjustment, the animation duration is no longer a fixed value, but an adaptive parameter that matches the actual reading needs of the text, thus avoiding the problem of animations being out of sync with the reading rhythm. Sub-step two: In teaching materials, text usually appears line by line or sentence by sentence. If the animations of two consecutive text units overlap in time or conflict in type, it will cause visual confusion for learners. For example, if the previous text unit is performing a fade-in animation lasting 1 second, and the next text unit starts a scaling animation at the same time, the superposition of these two animations will be dazzling. A more reasonable approach is to start the next animation only after the previous one has finished, or to change one of the animation types to a visually gentler one. To address this issue, this embodiment obtains the animation information of the adjacent text units in the original teaching text for the current text unit. Adjacent text units typically refer to the previous text unit, as the animation for the next text unit has not yet been generated. The system reads the animation configuration of the previous text unit, obtaining its animation type, start time, and end time. Then, based on this information, it adjusts the start time and type of the current text unit's animation. The rule for adjusting the start time is: if the end time of the previous animation is later than the currently preset start time, the start time of the current animation is postponed until after the previous animation ends, with a small interval, such as 0.1 seconds, added to ensure a clear separation between the two animations.The rules for adjusting animation types are as follows: a predefined list of conflicting types is used. For example, simultaneous fade-in and zoom can cause visual interference, and simultaneous blink and movement can also be distracting. If the type of the current animation conflicts with the type of the previous animation, the system will replace the current animation type with an alternative type. The alternative type can be selected from the template library as a semantically similar but non-conflicting type, such as changing zoom to slide-in. Through this adjustment, multiple animations on the same page can be presented in an orderly and harmonious sequence, maintaining the continuity of the teaching rhythm while avoiding visual clutter. Sub-step three: After the adjustments in the first two sub-steps, the three key attributes in the initial animation parameters—duration, start time, and type—have been modified to values ​​that adapt to the context. Other parameters that have not been adjusted, such as color, direction, and easing curve, retain their original values. These parameters are integrated into a structured data object, which is the animation configuration file. The animation configuration file is usually stored in JSON format and contains all the information required for rendering, which can be directly read and executed by the subsequent rendering module. It should be noted that the generation of the animation configuration file does not mean the end of the parameter adjustment process. In practice, the animation configuration file can contain other parameters that need to be adjusted, such as the position offset of the target object and the specific coordinates of the movement path.

[0021] In this embodiment, traditional methods focus only on individual text units when generating animations, neglecting the contextual relationships between texts and the reasonable constraints of text length on animation duration. This results in a large number of samples in the generated dataset where animation duration does not match the reading rhythm, and samples where preceding and following animations interfere with each other. These samples are unusable in actual teaching. This method solves the above problems through two specific adjustment operations. The first operation uses text length features to adjust the animation duration, ensuring short texts are paired with short animations and long texts with long animations, and providing boundary protection in extreme length cases to ensure reasonable duration. The second operation uses the animation information of adjacent texts to adjust the start time and type, preventing time overlap or visual conflict between animations and ensuring smooth presentation.

[0022] In one embodiment, the steps of labeling and data augmenting the paired samples include: Extract text units, motion effect configuration files, and dynamic videos from the paired samples, and generate labeled data according to a preset field structure. The field structure includes text content, motion effect type, motion effect duration, motion effect start time, target object, and video storage path. The text unit is rewritten, and the rewritten text unit is associated with and stored with the motion effect configuration file and dynamic video in the paired sample to generate the first enhanced sample. At least one parameter in the motion effect configuration file of the paired sample is numerically perturbed, the perturbed motion effect configuration file is re-rendered to generate a new dynamic video, and stored in association with the original text unit to generate a second enhanced sample.

[0023] As described above, sub-step one has already obtained a batch of paired samples that have passed the consistency check through the aforementioned embodiments. Each sample contains text units, animation configuration files, and dynamic videos. However, these samples themselves are just discrete files or data objects, lacking unified descriptive information, which is not conducive to subsequent retrieval, filtering, and model training. Therefore, annotation is required to add structured metadata to each sample. The specific method of annotation is to extract key information from the paired samples and organize it according to a preset field structure. The preset field structure includes at least the following six fields. The first field is the text content, that is, the original teaching text itself, such as the Earth's orbit around the sun being approximately elliptical. The second field is the animation type, such as fade-in, slide-in, blink, or zoom. The third field is the animation duration, recording the duration of the animation from start to finish in seconds. The fourth field is the animation start time, usually expressed as an offset relative to the time when the text appears, such as 0 seconds indicating immediate start, and 0.5 seconds indicating a half-second delay. The fifth field is the target, indicating which noun phrase or region in the text the animation specifically applies to, such as formulas or... Figure 2The curve in the image. The sixth field is the video storage path, pointing to the location of the dynamic video file corresponding to the paired sample on the disk or cloud storage. These fields are organized into a record, and the common storage format is JSONL, that is, one JSON object per line. Each JSON object corresponds to a paired sample, containing key-value pairs of the above six fields. Through such annotation, the originally isolated sample files can be transformed into a well-structured, searchable, and statistically valid dataset. Researchers can filter all flashing samples based on the motion effect type, or find samples containing specific keywords based on the text content, which facilitates the subsequent use of the dataset. After the annotation in sub-step two is completed, the dataset has basic usability, but the number of samples may still be limited. In order to expand the scale of the dataset and enhance its diversity, data augmentation is required. This embodiment provides two augmentation methods. The first is to rewrite the text units. Text rewriting refers to replacing text units with synonyms or changing sentence structure while keeping the original meaning unchanged. For example, the original text "Pay attention to this formula" can be rewritten as "Please pay attention to the formula here" or "Everyone should pay attention to this mathematical expression". The rewritten text units are semantically equivalent to the original text units, but differ in word choice and sentence structure. This rewriting can be done through rule replacement using a thesaurus or automatically generated using pre-trained language models such as T5 or GPT. After rewriting, the rewritten text units are associated with and stored in the motion effect configuration files and dynamic videos of the original paired samples. In other words, the new text units reuse the original motion effect configuration and video to form a new paired sample. This is on the premise that the rewriting does not change the core semantics of the text, and the original motion effects are still appropriate for the rewritten text. For example, both "note this formula" and "please pay attention to this formula" express the intention of emphasis, and the blinking motion effect is applicable to both. In this way, each original sample can be expanded into multiple variant samples, and the size of the dataset can be multiplied. In sub-step three, the second enhancement method is to numerically perturb the parameters in the motion effect configuration file, rather than modifying the text. The motion effect configuration file contains multiple numerical parameters, such as motion effect duration, movement distance, rotation angle, scaling ratio, control points of the easing curve, etc. Applying small random changes to these parameters can generate visually similar but different motion effects. The specific method involves selecting at least one parameter from the motion effect configuration file, such as the motion effect duration, and multiplying it by a random coefficient between 0.8 and 1.2 to obtain a new duration value. Alternatively, a random offset within ±5 degrees can be added to the movement direction. The range of the perturbed parameters needs to be controlled within a reasonable range to avoid disrupting the basic semantics of the motion effect. For example, perturbing the fade-in duration from 0.5 seconds to 1.2 seconds still results in a reasonable fade-in effect, but perturbing it to 5 seconds would appear abnormally slow; therefore, upper and lower limits need to be set. After perturbing, a new dynamic video is generated by re-rendering using the modified motion effect configuration file.The new video still semantically matches the original text unit because the type and target of the motion effect remain unchanged; only the parameters have been slightly adjusted. The new video is then associated with and stored with the original text unit to form a second augmented sample. The benefit of parameter perturbation is that it generates a large number of visually rich but semantically consistent motion effect variants, helping to improve the model's robustness to minor changes in motion effect parameters.

[0024] In one embodiment, before the steps of identifying the sentence skeleton of the text unit, determining the semantic function type of the text unit based on the identification result, and extracting preset motion trigger keywords from the text unit, the method further includes: Detect whether a preset pair of conjunctions exists in the text unit; If a preset pair of conjunctions exists in the text unit, the text unit is split into multiple clauses according to the pair of conjunctions; Based on the results of the splitting, the sentence structure skeleton of the clause is identified, the semantic function type of the clause is determined, and the motion effect triggering keywords in the clause are extracted. Combining the semantic function type and the motion effect triggering keywords, the corresponding motion effect type and motion effect intensity are found from the preset mapping rules. Analyze the temporal meaning of the connecting word pairs, and determine the temporal relationship between the motion effects corresponding to the multiple clauses based on the temporal meaning. The temporal relationship includes a serial relationship or a parallel relationship. The motion effect type, motion effect intensity, and target of each clause are combined with the temporal relationship to generate a composite motion effect semantic tag set, which replaces the motion effect semantic tags of the text unit.

[0025] As mentioned above, before formally parsing the sentence structure of a text unit, sub-step one first needs to determine whether the text unit is a complex sentence. Complex sentences are usually formed by conjunctions connecting two or more clauses, which imply the logical relationship between the clauses. Common conjunction pairs in teaching texts include firstly…then, on the one hand…on the other hand, not only…but also, although…however, etc. These conjunction pairs can be pre-organized into a list and stored in the system. The specific detection method is to scan the string content of the text unit and see if it matches the regular expression pattern of any conjunction pair in the list. For example, if the text contains "firstly" and "then" appears shortly afterward, it is determined that the conjunction pair "firstly…then" exists. If the pre-defined conjunction pair is detected, it means that the current text unit contains multiple independent motion intentions and needs to enter the subsequent splitting process. If no conjunction is detected, the entire text unit is processed directly according to the original process. Sub-step two, after detecting the conjunction pair, needs to split the text unit into multiple independent clauses. The splitting is based on the two conjunctions in the conjunction pair as the dividing point. For example, given the text "First, observe the trend of the chart, then compare the two sets of data," the system will split the text into two clauses: "Please observe the trend of the chart" and "Compare the two sets of data," using "first" and "then" as the boundaries. When splitting, it's important to preserve the integrity of each clause. Removing the conjunction itself or incorporating it into the preceding clause are viable strategies, depending on the conjunction's position within the sentence. For conjunction pairs like "on the one hand…on the other hand," the two resulting clauses are usually parallel, each expressing independent semantics. After splitting, a complex sentence becomes two or more independent clauses. Each clause is shorter, with more focused semantics, facilitating subsequent semantic analysis. The split clauses are numbered according to their order of appearance in the original text to determine the temporal relationship of the animation. After obtaining the clause list in sub-step three, the system needs to independently perform the original semantic analysis operation on each clause. Specifically, for each clause, the system identifies its sentence structure skeleton, determines its semantic function type, extracts the animation trigger keywords, and then searches for the corresponding animation type and intensity from the mapping rules based on the combination of semantic function type and keywords. Simultaneously, the system extracts the target object from the clauses, i.e., the noun phrases in the clause that need to be emphasized or displayed in the animation. This operation means that each clause generates its own animation type, intensity, and target object. For example, the first clause, "Please observe the trend of the chart," might be mapped to a directional moving animation with medium intensity because it contains the keyword "observe," and its target object is the trend of the chart. The second clause, "Compare two sets of data," might be mapped to a left-right split-screen comparison animation with medium intensity because it contains the keyword "compare," and its target object is both sets of data. In this way, the previously ignored intention of the second animation is fully expressed. In sub-step four, different conjunction pairs imply different temporal meanings.Some conjunction pairs indicate sequential relationships, such as "first...then," "first...then," and "first step...second step." These conjunction pairs imply the order in which actions occur, and the corresponding animations should be played sequentially, meaning the animation of the first clause should be played before the animation of the second clause. Other conjunction pairs indicate parallel relationships, such as "on the one hand...on the other hand" and "while...while." These conjunction pairs imply that actions occur simultaneously or in any order, and the corresponding animations can be played in parallel, meaning the animations of the two clauses can be displayed simultaneously within the same time period. This method needs to determine the temporal relationship based on the type of conjunction pair. This can be achieved through a predefined mapping table that records the temporal relationship type corresponding to each conjunction pair. For "first...then," the temporal relationship is determined to be sequential. For "on the one hand...on the other hand," the temporal relationship is determined to be parallel. For some semantically ambiguous conjunction pairs, a sequential relationship can be used by default to ensure clarity of display. Finally, in sub-step five, the animation type, animation intensity, and target of each clause are combined with the previously determined temporal relationship to generate a composite animation semantic tag set. This composite tag set is a structured data object containing multiple sub-tags, each corresponding to a clause, and the entire tag set carries a temporal relationship attribute. The purpose of the composite tag set is to replace the original single animation semantic tag for the entire text unit. In subsequent animation parameter matching and rendering steps, the system sees the composite tag set and knows that multiple animations need to be generated, and these animations must be arranged according to the specified temporal relationship. For sequential relationships, the rendering engine will play the animation of the first clause first, and then play the animation of the second clause after that. For parallel relationships, the rendering engine will play multiple animations simultaneously and overlay them on the corresponding text areas. In this way, the multiple animation intentions in what was originally a composite sentence are expressed and implemented completely and logically.

[0026] It is worth mentioning that in teaching texts, complex sentences introduced by conjunctions frequently appear, such as "First pay attention to this formula, then compare..." Figure 2 and Figure 3 The first part of this sentence requires an emphasis-type motion effect, and the second part requires a contrast-type motion effect. If conventional semantic parsing methods are used, the entire sentence can only be assigned one motion effect tag—either emphasis or contrast—and the other motion effect intention will be discarded. This approach cannot fully express the multiple motion effect intentions in a complex sentence, nor can it reflect the sequential or simultaneous relationship between two motion effects.

[0027] This embodiment solves the aforementioned problems by introducing a connector pair detection and clause splitting mechanism. First, it scans the text units for pre-defined connector pairs, such as "firstly...then" or "on the one hand...on the other hand." Once detected, the original text is split into multiple independent clauses according to the position of the connector. Next, a complete semantic analysis is performed on each clause, independently matching the animation type, intensity, and target object for each clause, ensuring that the animation intent of each clause is not overlooked. Then, the temporal meaning implied by the connector pairs is analyzed to determine whether these animations should be played sequentially or in parallel. Finally, the animation information and temporal relationships of all clauses are combined into a composite animation description. In this way, multiple animation intents that would otherwise be discarded are fully preserved, and the logical order between animations is accurately expressed. During subsequent rendering, these animations are played sequentially or simultaneously according to the requirements of sequential or parallel execution, ensuring that the teaching animation display of the compound sentence conforms to the semantic structure of the original text.

[0028] In one embodiment, the step of adjusting the motion effect duration in the initial motion effect parameters according to the length feature includes: Obtain the number of valid characters in the text unit; When the number of valid characters is less than or equal to the first threshold, the animation duration is determined as the first preset duration, wherein the first preset duration is the minimum renderable duration supported by the animation template corresponding to the initial animation parameters; When the number of valid characters is greater than or equal to the second threshold, the animation duration is determined as the second preset duration, which is equal to the maximum renderable duration supported by the animation template, and the second threshold is greater than the first threshold. When the number of valid characters is greater than the first threshold and less than the second threshold, the baseline duration is calculated based on the product of the number of valid characters and the reading time per unit character, and the baseline duration is used as the animation duration.

[0029] As mentioned above, the first step in adjusting the duration is to determine the actual length of the text unit. The effective character count used here is not simply the total number of characters, but rather the number of characters that students actually need to read after removing punctuation, spaces, and meaningless decorative symbols. For example, a text unit might be "Please note, Figure 2The curve in the text is clearly rising! Exclamation marks and commas are not counted as valid characters. Every Chinese character in "Please note," the number 2, and words like "curve," "clearly," and "rising" are counted as valid characters. Special symbols commonly used in teaching scenarios, such as superscripts and subscripts in mathematical formulas and arrows, also have semantic information and should be counted as valid characters. The calculation of valid characters can be based on a pre-defined rule, for example, each Chinese character counts as 1, each number and English letter counts as 1, punctuation marks are excluded, and special symbols are weighted according to their importance. When the number of valid characters is very small, such as only one character or one exclamation mark, the baseline time might be only 0.3 seconds or even less using the conventional linear calculation method. However... Each animation template has a minimum renderable duration; below this duration, the animation cannot be displayed correctly. For example, a fade-in animation requires at least 0.5 seconds to complete the transition from completely transparent to completely opaque, and a blinking animation requires at least one complete blink cycle, approximately 0.6 seconds, for a single blink to be perceptible. If the duration is forcibly set too short, the animation will flash by or even fail to trigger, rendering the generated video useless in actual teaching. Therefore, this embodiment sets a first threshold, typically 2 to 3 characters. When the number of valid characters in a text unit is less than or equal to this first threshold, the system no longer uses a linear calculation formula but directly sets the animation duration to the duration supported by the currently matched animation template. The minimum renderable duration is set. For example, for a text unit containing only the word "good," the effective character count is 1, which is less than the first threshold. The system will then query the minimum renderable duration of the fade-in template, which is 0.5 seconds, and set the animation duration to 0.5 seconds. This ensures that even very short texts can be fully displayed without flashing by. Conversely, when text units are very long, such as a theorem definition containing hundreds of words, the linearly calculated baseline duration may reach 10 seconds or even longer. However, longer durations are not always better for animations. Some types of animations become abnormally slow or lose their rhythm after a certain duration. For example, a blinking animation will become annoying for students if it blinks continuously for more than 3 seconds; a slide-in animation will also become annoying if it blinks too slowly. A duration exceeding 5 seconds can feel sluggish. Each animation template has a reasonable maximum duration; exceeding this limit severely degrades the visual effect or even has the opposite effect. Therefore, this embodiment sets a second threshold, which is greater than the first threshold and typically ranges from 50 to 80 characters. When the number of valid characters in a text unit is greater than or equal to this second threshold, the system directly sets the animation duration to the maximum renderable duration supported by that animation template. For example, for a long paragraph of 100 characters, the number of valid characters far exceeds the second threshold, so the system will set the animation duration to the maximum renderable duration of the blinking template, 3 seconds, instead of the linearly calculated 10 seconds. This ensures that the animation does not appear sluggish due to excessive length.When the number of effective characters in a text unit falls between the first and second thresholds—that is, when the text is neither extremely short nor extremely long—the system uses a linear calculation formula to determine the animation duration. The specific formula is: the baseline duration equals the number of effective characters multiplied by the reading time per character. The reading time per character is a configurable parameter, typically set based on the average reading speed of students in a teaching scenario, for example, 0.25 to 0.35 seconds per character. Assuming the number of effective characters is 20 and the reading time per character is 0.3 seconds, the baseline duration is 6 seconds. This duration is directly used as the animation duration without further boundary trimming. For most normal-length teaching texts, this linear calculation method effectively matches the students' reading pace, ensuring the animation ends shortly after the text is read, creating a smooth reading experience.

[0030] It should be noted that the first and second preset durations in this embodiment are not fixed global values, but are associated with the currently matched animation template. Different animation templates have different minimum and maximum renderable durations; for example, the minimum duration for a fade-in template might be 0.5 seconds, while the minimum duration for a blink template might be 0.6 seconds. This embodiment determines the boundary values ​​by querying the template's own attributes, allowing the duration adjustment to adapt to the characteristics of different animation types.

[0031] In one embodiment, after the steps of identifying the sentence structure skeleton of the text unit, determining the semantic function type of the text unit based on the identification result, and extracting preset motion effect triggering keywords from the text unit, the method further includes: When no motion effect triggering keyword is extracted from the text unit and the semantic function type is a declarative sentence, noun phrases are extracted from the text unit as the objects to be acted upon, and the preset default motion effect type and default motion effect intensity are used as the found motion effect type and motion effect intensity. Skip the step of searching for the corresponding motion effect type and intensity from the preset mapping rules.

[0032] As mentioned above, not all sentences in teaching texts contain obvious animation trigger keywords. For example, a purely declarative sentence like "The Earth's orbit around the Sun is approximately elliptical" lacks prompts like "note," "first step," or "in short," and is neither an interrogative nor an exclamatory sentence, but a plain, objective statement. Following conventional semantic parsing, due to the lack of keywords, the system cannot find the corresponding animation type and intensity from the mapping rules, potentially resulting in a null value or an inappropriate default animation, making it impossible to obtain a reasonable animation pairing for such text. This embodiment adds a judgment branch after completing sentence skeleton recognition and keyword extraction. The system checks two conditions: first, no preset animation trigger keywords are extracted from the text unit; second, the semantic function type of the text unit is determined to be a declarative sentence. When both conditions are met, it indicates that the current text is ordinary narrative content without any special animation intention. For this type of text, this embodiment no longer attempts to search from the mapping rules but instead enters a dedicated default processing flow. Since there are no keywords to indicate which specific object the animation should apply to, it's necessary to automatically identify the most worthy content to be emphasized in the animation from the text. A common approach is to extract noun phrases from the text as the objects to be animated. These noun phrases can be the subject, object, or core noun components of a sentence. For example, for the sentence "The Earth's orbit around the sun is approximately elliptical," the system can identify noun phrases such as "Earth," "sun," "orbit," and "elliptical" through dependency parsing. Further, the most core noun phrase, such as "orbit," or the entire subject can be selected. A specific selection strategy could be to choose the first noun phrase appearing in the text, or the noun with the highest weight. The purpose of extracting noun phrases is to provide a clear target for the subsequent default animation, ensuring the animation isn't left unplaced. After determining the object, a default animation type and intensity need to be assigned to this text unit. The default animation type is usually a gentle animation suitable for most scenarios, such as a fade-in animation. A fade-in effect that isn't overly flashy, yet makes the text appear natural and orderly, is ideal for straightforward narrative content. The default fade-in intensity is generally set to low to avoid excessive animation interfering with students' reading. Specifically, a low-intensity fade-in effect might appear as a gradual transition from 0.2% transparency to full visibility over 0.5 seconds, rather than a sudden appearance from complete transparency. After assigning default values, the original step of "searching for the corresponding animation type and intensity from the preset mapping rules" should be explicitly skipped. This is because the default values ​​have already replaced the search result; continuing the search might result in an error due to the lack of a matching rule.

[0033] In one feasible embodiment, during the text rewriting and enhancement process, the system performs synonym replacement or sentence transformation on the original text units, and then directly associates and stores the rewritten text with the motion effect configuration file and dynamic video in the original sample. This method implicitly assumes that all parameters in the original motion effect configuration, especially the motion effect duration, are still applicable to the rewritten text. However, in practice, the rewriting operation often causes a significant change in the text length.

[0034] refer to Figure 5 Let's take a specific scenario as an example: The original text is "Attention.", which has only 2 effective characters. According to the rules for processing very short texts, the animation duration is set to 0.5 seconds. After rewriting the text, we get "Please pay special attention to this important knowledge point," increasing the effective character count to 15. According to the normal text duration calculation rules, 15 characters multiplied by the reading time of 0.3 seconds per character, the reasonable animation duration should be 4.5 seconds. However, if the original 0.5-second animation configuration is directly reused, the animation ends in a very short time, while students need several seconds to read the rewritten long text, causing a serious mismatch between the animation and the reading rhythm. Conversely, if the original text is long and the rewritten version is short, there will be an awkward situation where the animation is dragging and students have already finished reading but the animation is still playing.

[0035] Therefore, the solution proposed in this embodiment includes: rewriting the text unit, and associating and storing the rewritten text unit with the motion effect configuration file and dynamic video in the paired sample, the steps of which include: Get the number of valid characters in the rewritten text cell; When the number of valid characters is less than or equal to the first threshold, the animation duration is determined as the first preset duration; when the number of valid characters is greater than or equal to the second threshold, the animation duration is determined as the second preset duration; when the number of valid characters is greater than the first threshold and less than the second threshold, a base duration is calculated based on the product of the number of valid characters and the reading time per unit character, and the base duration is used as the animation duration. Update the animation duration parameter in the animation configuration file of the paired sample with the animation duration; The rewritten text unit is associated with and stored in conjunction with the updated animation configuration file and the dynamic video.

[0036] As mentioned above, after rewriting the original text unit, a completely new text unit is obtained. This rewritten text is semantically identical to the original, but the wording and sentence structure have changed, so its length may differ significantly from the original. For example, the original text "Note" has only two characters; after being rewritten as "Please pay special attention to this important knowledge point," the effective character count increases from 2 to 15. If the animation duration corresponding to the original text is continued, for example, 0.5 seconds, then when the rewritten long text appears, the animation ends within 0.5 seconds, while students may need more than 3 seconds to read it, creating a noticeable disconnect. Therefore, this embodiment first obtains the effective character count of the rewritten text before associating it with the original animation configuration file. The calculation rule for the effective character count is the same as before: punctuation marks and spaces are excluded, and only Chinese characters, numbers, letters, and meaningful special symbols are counted. The purpose of obtaining this value is to provide a basis for subsequent duration recalculation. After obtaining the effective character count of the rewritten text, the rewritten text is treated as a new independent text unit, and the appropriate animation duration is recalculated based on its length characteristics. Specific rules are divided into three cases: When the effective character count of the rewritten text is less than or equal to the first threshold, it indicates that the text is very short, possibly containing only one word or a short sentence. In this case, the animation duration is set to the first preset duration, which is the minimum renderable duration supported by the current animation template, to ensure the animation can be displayed completely. When the effective character count is greater than or equal to the second threshold, it indicates that the text is very long. In this case, the animation duration is set to the second preset duration, which is the maximum renderable duration supported by the animation template, to avoid the animation being too sluggish. When the effective character count is between the first and second thresholds, a baseline duration is calculated by multiplying the effective character count by the reading time per unit character, and this baseline duration is used as the animation duration. Through this recalculation, the animation duration of the rewritten text is no longer the duration of the original text, but a new duration that matches its own length. For example, the original text "Attention" has only 2 characters and a duration of 0.5 seconds. The rewritten text has 15 characters, and at 0.3 seconds per character, the duration is 4.5 seconds. Since 4.5 seconds may be less than the second threshold, the new duration is 4.5 seconds. This gives the longer text a sufficiently long animation display time, realigning the reading rhythm with the animation rhythm. After recalculating the appropriate animation duration, the animation configuration file in the original paired sample needs to be modified. Specifically, the animation duration field in the configuration file is located, and its original value is replaced with the newly calculated value. Other parameters in the configuration file, such as animation type, start time, target, and easing curve, remain unchanged. This partial update method ensures the semantic consistency of the animation and corrects the duration mismatch problem. Finally, the rewritten text unit, the updated animation configuration file, and the dynamic video in the original paired sample are stored together to form a new enhanced sample.It's important to note that the associated dynamic video here is still the original video file. There might be slight inconsistencies between the updated configuration file and the original video because the video is rendered according to its original duration, while the duration in the configuration file has been modified. To completely resolve the consistency issue, a more robust solution is to re-render and generate a new dynamic video after updating the duration. This embodiment only requires the original video to be associated and stored, but in actual implementation, re-rendering can be chosen as needed. Regardless of the method, at least the duration parameters in the text and configuration file are guaranteed to match, allowing for subsequent re-rendering of the correct video based on the updated configuration file.

[0037] Reference Figure 3 This application also provides an automated system for constructing a teaching text animation matching dataset, including: Module 1 is used to acquire raw teaching text data and preprocess the raw teaching text data to obtain multiple text units; The parsing and recognition module 2 is used to perform semantic parsing on each text unit, identify the motion effect triggering information in the text unit, and generate a corresponding motion effect semantic tag set based on the recognition result. The motion effect triggering information is a text feature that can be mapped to the motion effect type, motion effect intensity, or motion effect target. Module 3 is invoked to call the pre-built motion effect parameterization template library and match the motion effect semantic tag set with the corresponding initial motion effect parameters from the motion effect parameterization template library; Analysis and generation module 4 is used to analyze the association features of the text unit, adjust the initial animation parameters according to the association features of the text unit, and generate an animation configuration file corresponding to the text unit; Storage module 5 is used to render the motion effect configuration file to generate a dynamic video, and to perform consistency verification between the dynamic video and the corresponding text unit, and to associate and store the text unit, motion effect configuration file and dynamic video that pass the verification as a paired sample. Enhancement processing module 6 is used to label and enhance the paired samples to generate a teaching text animation pairing dataset.

[0038] As described above, it is understood that each component of the automated construction system for the teaching text animation matching dataset proposed in this application can realize the function of any of the automated construction methods for the teaching text animation matching dataset described above, and the specific structure will not be repeated.

[0039] Reference Figure 4 This application also provides a computer device, which may be a server, and its internal structure may be as follows: Figure 4As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores monitoring data and other data. The network interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements an automated method for constructing a teaching text-to-animation matching dataset.

[0040] The processor described above executes the automated construction method for the teaching text animation pairing dataset, including: acquiring raw teaching text data; preprocessing the raw teaching text data to obtain multiple text units; performing semantic parsing on each text unit to identify animation trigger information in the text unit, and generating a corresponding animation semantic tag set based on the identification result; calling a pre-built animation parameterization template library to match the animation semantic tag set with the corresponding initial animation parameters from the animation parameterization template library; analyzing the association features of the text units, adjusting the initial animation parameters based on the association features of the text units, and generating an animation configuration file corresponding to the text unit; rendering the animation configuration file to generate a dynamic video, and performing consistency verification between the dynamic video and the corresponding text unit, and storing the verified text units, animation configuration files, and dynamic videos as paired samples; and annotating and data augmenting the paired samples to generate a teaching text animation pairing dataset.

[0041] One embodiment of this application also provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements an automated method for constructing a teaching text-motion matching dataset, including the following steps: acquiring original teaching text data; preprocessing the original teaching text data to obtain multiple text units; performing semantic parsing on each text unit to identify motion effect triggering information in the text unit, and generating a corresponding motion effect semantic tag set based on the identification result; calling a pre-built motion effect parameterization template library to match the motion effect semantic tag set with corresponding initial motion effect parameters from the motion effect parameterization template library; analyzing the association features of the text units, adjusting the initial motion effect parameters based on the association features of the text units, and generating a motion effect configuration file corresponding to the text unit; rendering the motion effect configuration file to generate a dynamic video, and performing consistency verification between the dynamic video and the corresponding text unit, and storing the verified text units, motion effect configuration files, and dynamic videos as matching samples; and labeling and data augmenting the matching samples to generate a teaching text-motion matching dataset.

[0042] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in this application and in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0043] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0044] The above description is only a preferred embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural changes made based on the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. An automatic construction method of teaching character motion effect pairing dataset, characterized in that, The method includes: Obtain the original teaching text data, and preprocess the original teaching text data to obtain multiple text units; Semantic parsing is performed on each text unit to identify motion effect triggering information in the text unit. Based on the identification results, a corresponding set of motion effect semantic tags is generated. The motion effect triggering information is a text feature that can be mapped to motion effect type, motion effect intensity, or motion effect target. Call the pre-built motion effect parameterization template library and match the motion effect semantic tag set with the corresponding initial motion effect parameters from the motion effect parameterization template library; Analyze the association features of the text unit, adjust the initial animation parameters according to the association features of the text unit, and generate an animation configuration file corresponding to the text unit; The motion effect configuration file is rendered to generate a dynamic video, and the consistency between the dynamic video and the corresponding text unit is checked. The text unit, motion effect configuration file and dynamic video that pass the check are associated and stored as a paired sample. The paired samples are labeled and augmented to generate a teaching text animation pairing dataset.

2. The method of claim 1, wherein, The steps of performing semantic parsing on each text unit, identifying motion effect triggering information in the text unit, and generating a corresponding motion effect semantic tag set based on the identification results include: Identify the sentence structure skeleton of the text unit, determine the semantic function type of the text unit based on the identification result, and extract preset animation trigger keywords from the text unit; Based on the combination of the semantic function type and the motion effect triggering keyword, the corresponding motion effect type and motion effect intensity are searched from the preset mapping rules; Based on the found motion effect type and motion effect intensity, generate motion effect semantic tags that include the motion effect type, motion effect intensity, and the object to be applied in the text unit.

3. The method of claim 2, wherein, Before the steps of identifying the sentence structure skeleton of the text unit, determining the semantic function type of the text unit based on the identification result, and extracting preset animation trigger keywords from the text unit, the method further includes: Detect whether a preset pair of conjunctions exists in the text unit; If a preset pair of conjunctions exists in the text unit, the text unit is split into multiple clauses according to the pair of conjunctions; Based on the results of the splitting, the sentence structure skeleton of the clause is identified, the semantic function type of the clause is determined, and the motion effect triggering keywords in the clause are extracted. Combining the semantic function type and the motion effect triggering keywords, the corresponding motion effect type and motion effect intensity are found from the preset mapping rules. Analyze the temporal meaning of the connecting word pairs, and determine the temporal relationship between the motion effects corresponding to the multiple clauses based on the temporal meaning. The temporal relationship includes a serial relationship or a parallel relationship. The motion effect type, motion effect intensity, and target of each clause are combined with the temporal relationship to generate a composite motion effect semantic tag set, which replaces the motion effect semantic tags of the text unit.

4. The method of claim 2, wherein, After the steps of identifying the sentence structure skeleton of the text unit, determining the semantic function type of the text unit based on the identification result, and extracting preset motion effect triggering keywords from the text unit, the method further includes: When no motion effect triggering keyword is extracted from the text unit and the semantic function type is a declarative sentence, noun phrases are extracted from the text unit as the objects to be acted upon, and the preset default motion effect type and default motion effect intensity are used as the found motion effect type and motion effect intensity. Skip the step of searching for the corresponding motion effect type and intensity from the preset mapping rules.

5. The method of claim 1, wherein, The steps of analyzing the association features of the text units, adjusting the initial animation parameters based on the association features of the text units, and generating an animation configuration file corresponding to the text units include: Obtain the length feature of the text unit, and adjust the animation duration in the initial animation parameters according to the length feature; Obtain the animation information of the adjacent text units of the text unit in the original teaching text, and adjust the animation start time and animation type in the initial animation parameters according to the animation information of the adjacent text units; The adjusted animation duration, animation start time, and animation type are integrated into the animation configuration file.

6. The method of claim 5, wherein, The step of adjusting the animation duration in the initial animation parameters according to the length feature includes: Obtain the number of valid characters in the text unit; When the number of valid characters is less than or equal to the first threshold, the animation duration is determined as the first preset duration, wherein the first preset duration is the minimum renderable duration supported by the animation template corresponding to the initial animation parameters; When the number of valid characters is greater than or equal to the second threshold, the animation duration is determined as the second preset duration, which is equal to the maximum renderable duration supported by the animation template, and the second threshold is greater than the first threshold. When the number of valid characters is greater than the first threshold and less than the second threshold, the baseline duration is calculated based on the product of the number of valid characters and the reading time per unit character, and the baseline duration is used as the animation duration.

7. The method of claim 1, wherein the method further comprises: The steps of labeling and data augmenting the paired samples include: Extract text units, motion effect configuration files, and dynamic videos from the paired samples, and generate labeled data according to a preset field structure. The field structure includes text content, motion effect type, motion effect duration, motion effect start time, target object, and video storage path. The text unit is rewritten, and the rewritten text unit is associated with and stored with the motion effect configuration file and dynamic video in the paired sample to generate the first enhanced sample. At least one parameter in the motion effect configuration file of the paired sample is numerically perturbed, the perturbed motion effect configuration file is re-rendered to generate a new dynamic video, and stored in association with the original text unit to generate a second enhanced sample.

8. An automated construction system of teaching character motion effect pairing dataset, characterized in that, include: The acquisition module is used to acquire raw teaching text data and preprocess the raw teaching text data to obtain multiple text units. The parsing and recognition module is used to perform semantic parsing on each text unit, identify the motion effect triggering information in the text unit, and generate a corresponding motion effect semantic tag set based on the recognition result. The motion effect triggering information is a text feature that can be mapped to the motion effect type, motion effect intensity, or motion effect target. The calling module is used to call the pre-built motion effect parameterization template library and match the motion effect semantic tag set with the corresponding initial motion effect parameters from the motion effect parameterization template library; The analysis and generation module is used to analyze the association features of the text unit, adjust the initial animation parameters according to the association features of the text unit, and generate an animation configuration file corresponding to the text unit. The storage module is used to render the motion effect configuration file to generate a dynamic video, and to perform consistency verification between the dynamic video and the corresponding text unit. The text unit, motion effect configuration file and dynamic video that pass the verification are associated and stored as a paired sample. The enhancement processing module is used to label and augment the paired samples to generate a teaching text animation pairing dataset. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images