An AI interaction method and system for instant generation of composition writing ideas

Abstract

The application belongs to the technical field of education auxiliary, and discloses an AI interaction method and system for instant generation of composition writing ideas, which comprises the following steps: collecting writing task related input data of a user and performing data cleaning to obtain optimized writing task input data; performing structured semantic analysis on the optimized writing task input data to obtain a composition task intention data set; performing cognitive degree matching on the composition task intention data set based on personal data of the user to output a user-adapted composition writing idea; performing emotion rhythm promotion on the user-adapted composition writing idea to generate an emotion visualized composition writing idea; performing semantic embedding on the emotion visualized composition writing idea and new writing requirements of the user to obtain a complete composition writing idea; and performing feedback evaluation on the complete composition writing idea and sending the evaluation result and the complete composition writing idea to a preset user terminal.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of education assistance, more particularly, the present application relates to an AI interaction method and system for generating writing ideas for compositions. BACKGROUND

[0002] In the field of Chinese education today, writing is an essential teaching content, but for students in lower grades, most students may have difficulty in independently constructing a reasonable composition writing idea; in the case of the gradual entry of artificial intelligence technology into the field of education today, automatically generating a composition writing idea that meets the user's requirements through a method combining artificial intelligence has become a key development direction in the field of Chinese education, but the existing composition idea generation method still has certain problems.

[0003] Since the user may give different types of data as the basis for generating a composition idea, the existing composition idea generation method is very rough in processing each type of data; when the input data is image data including composition requirement content, the existing composition idea generation method may call an OCR model to recognize the text therein, but there may still be certain recognition errors, resulting in semantic divergence between the generated characters and other characters, and logical errors in the generated idea; when the input data is pure image data, the existing composition idea method can recognize the things in the image, but it is difficult to distinguish the importance of each thing in the composition idea, which may cause the attention to be focused on certain irrelevant things in the image, and the direction of the composition idea may be deviated by ignoring factors such as character behavior and key events; when the input data is the user's voice input data, the existing composition idea generation method may only focus on the accuracy of speech recognition, accurately transcribing the text content, but it ignores logical adjustment of the text content, resulting in an incorrect composition writing idea; for example, a certain user's sentence may have a logical problem, but the existing composition idea generation method only transcribes the voice content of the sentence completely without any adjustment, resulting in a wrong writing intention direction of the subsequently generated composition idea, which cannot form a reasonable logic; at the same time, the existing composition idea generation method does not take into account the user's cognitive level, and may generate a composition writing idea composed of complex logical structures and abstract writing elements, resulting in cognitive load on the user and indirectly reducing the user's learning enthusiasm; in addition, the existing composition idea generation method generally lacks analysis of emotional changes in the composition idea, making it difficult for the user to understand the rhythm of emotional changes and grasp the level of emotional expression, reducing the teaching efficiency; on the other hand, it is a normal behavior for the user to supplement temporary requirements during the writing process, and the existing composition idea generation method often directly adds the supplemented temporary requirements to the end of the currently generated content, which may cause logical and semantic disconnection and cause logical discontinuity.

[0004] In view of this, the present application proposes an AI interaction method and system for instant generation of composition writing ideas to solve the above problems. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, in order to achieve the above-mentioned purpose, the present application provides the following technical scheme: an AI interaction method for instant generation of composition writing ideas, comprising:

[0006] S1. Collecting writing task related input data of the user and performing data cleaning to obtain optimized writing task input data;

[0007] S2. Performing structured semantic analysis on the optimized writing task input data to obtain a composition task intention data set;

[0008] S3. Obtaining user personal data, performing cognitive degree matching on the composition task intention data set based on the user personal data, and outputting a user-adapted composition writing idea;

[0009] S4. Proceeding with the emotional rhythm of the user-adapted composition writing idea, and generating an emotional visual composition writing idea;

[0010] S5. Real-time collection of user's new writing requirements, semantic embedding of the emotional visual composition writing idea and the user's new writing requirements, and obtaining a complete composition writing idea;

[0011] S6. Feedback evaluation of the complete composition writing idea, and sending the evaluation result and the complete composition writing idea to a preset user terminal.

[0012] Further, the structured semantic analysis includes:

[0013] The data types of the optimized writing task input data include composition requirement image data, composition material image data and user voice input data;

[0014] The image data of the composition is subjected to gray scale normalization processing to obtain a character image unit; contour fitting is performed on the character image unit to generate character codes, and the character codes are arranged as character code segments; the number of pixel blocks corresponding to each character in the character code segment is detected and the pixel density of the corresponding position is calculated, and if the pixel density is less than a preset readable density threshold, a low readability label is added to the character; the glyph components of the character corresponding to the low readability label are recognized and candidate characters are constructed, and the candidate characters and the characters adjacent to the corresponding position are combined to form a candidate word vector; the semantic similarity of the candidate word vector and the content word vector adjacent to the position of the candidate word vector is calculated, and the candidate word vector with the highest semantic similarity is selected as the appropriate word vector and covers the original candidate word vector; the subject-predicate-object word phrases of each sentence in the adjusted character code segment are recognized and sentence node chains are constructed; the semantic relationship between the sentences is recognized, the sentence node chains are connected based on the semantic relationship, and a composition logical path graph is generated;

[0015] The image data of the composition material is subjected to label weight adjustment to obtain label hierarchical thinking data; the intention structure of the user voice input data is extracted to obtain structure forming thinking data; and any one of the composition logical path graph, the label hierarchical thinking data and the structure forming thinking data is selected as the composition task intention data set based on the data type.

[0016] Further, the label weight adjustment includes:

[0017] The semantic object in the composition material image data is recognized, and the edge contour of each semantic object is extracted; pixel segmentation is performed based on the edge contour of the semantic object to generate an image mask area of each semantic object, and a semantic label is added to the corresponding image mask area; the ratio of each image mask area to the area of the composition material image data is calculated to obtain a pixel area ratio; a boundary contour complexity function is constructed based on the area of the image mask area, and the boundary contour complexity of the image mask area is calculated using the function; the distance between the center coordinates of each image mask area and the center coordinates of the composition material image data is calculated to obtain a center distance; the pixel area ratio, the boundary contour complexity and the center distance are weighted and fused to obtain a comprehensive semantic weight of the corresponding image mask area; all image mask areas are sorted based on the comprehensive semantic weight to obtain a semantic priority sequence; the semantic priority sequence is matched with a preset image writing composition template to obtain label hierarchical thinking data.

[0018] Further, the intention structure extraction includes:

[0019] The user voice input data is converted into voice text content, a pause interval is detected, and the voice text content is divided into sentences based on the pause interval to obtain voice sentence content; the voice sentence content is processed by word segmentation, and each word segmentation is converted into a word segmentation vector; and the word segmentation vectors belonging to the same voice sentence content are combined into a multi-dimensional semantic representation vector;

[0020] The semantic distance of any two multi-dimensional semantic representation vectors is calculated, the multi-dimensional semantic representation vectors with a semantic distance less than a preset semantic distance threshold from any one multi-dimensional semantic representation vector are integrated into a high-density vector set, and the corresponding sentence region of the high-density vector set is identified; the semantic center vector of each high-density vector set is identified and a semantic center sequence is constructed; the angle change amount of adjacent semantic center vectors in the semantic center sequence is determined to judge the semantic change direction, and a semantic center change path is constructed; the position with an angle change amount higher than a preset angle threshold in the semantic center change path is identified, and the position is determined as a structural turning position; the sentence regions corresponding to the high-density vector sets before and after the structural turning position are reordered according to language logic to obtain structural forming idea data.

[0021] Further, the cognitive degree matching manner comprises:

[0022] The user personal data comprises historical composition text, historical composition score, user grade identifier and average writing time length; the number of sentences, the average sentence length and the number of sentence structure types of any one historical composition text in the historical composition text are counted, and a sentence complexity parameter group is constructed; the ratio of the number of functional words to the total number of words in the historical composition text is calculated to obtain a limited word density; the occurrence frequency of semantic relationship text in the historical composition text is identified; the sentence complexity parameter group, the limited word density and the occurrence frequency are dimensionally normalized to obtain a language structure parameter group, and the sum and average value are calculated to obtain a user average language structure parameter group; It should be noted that the acquisition of user personal data has a disclaimer: the user's authorization or consent to use user personal data is required;

[0023] The composition score standard deviation is calculated based on the historical composition score, and the ratio of the average composition score to the highest composition score is calculated to obtain a score fluctuation deviation; the absolute value of the difference between the average writing time length of the current user and the standard writing time length of students in the same grade is calculated based on the user grade identifier to obtain a time length deviation; the user cognitive ability vector is constructed based on the user average language structure parameter group, the score standard deviation, the score fluctuation deviation and the time length deviation, and the user grade identifier is used as the vector label;

[0024] extract writing elements structure, plot advancing step number and logical connection type number of the composition task intention dataset, and construct a composition intention complexity vector; compare the user cognitive ability vector with the composition intention complexity vector to output a cognitive bias index, adjust the composition task intention dataset based on the cognitive bias index, and output a user-adapted composition writing guideline.

[0025] Further, the output cognitive bias index includes:

[0026] The cognitive ability of the user cognitive ability vector is evaluated, and a cognitive ability level is output. The complexity of the composition intention complexity vector is estimated, and a task structure level is output. The difference between the cognitive ability level and the task structure level is calculated, which is the cognitive bias index;

[0027] The intention adjustment includes:

[0028] If the cognitive bias index is greater than zero, no processing is performed, otherwise the composition task intention dataset is simplified to obtain a user-adapted composition writing guideline.

[0029] Further, the emotion rhythm advancing includes:

[0030] The user-adapted composition writing guideline is divided into a guideline content unit, and emotional description content in each guideline content unit is identified. The emotion type of the emotional description content is detected and the emotion intensity is output by matching with a preset emotion level table. The guideline content unit is taken as a node and the emotion type is taken as a label, and an emotion intensity fluctuation curve is drawn by combining the structure order of the guideline content unit in the user-adapted composition writing guideline. The fluctuation slope of the emotion intensity fluctuation curve is calculated, and the node with a fluctuation slope higher than a preset slope threshold is marked as an emotion mutation node. The node with the highest emotion intensity in the emotion intensity fluctuation curve is identified as an emotion climax node. The nodes in the curve section after the emotion climax node are detected, which have an emotion intensity change amount higher than a preset intensity change threshold and a negative slope with the previous node, as emotion turning points. The emotion mutation node, the emotion climax node, the emotion turning point and the emotion type are mapped back to the guideline content unit and the corresponding index is added to obtain an emotion visualized composition writing guideline.

[0031] Further, the semantic embedding includes:

[0032] The user's new writing requirement is converted into text content, and sentence division is performed based on grammatical pauses to obtain a new requirement sentence; the subject in the new requirement sentence is extracted, and it is judged whether the subject is consistent with the subject of the emotion visualized writing thinking; if not, it is determined that semantic embedding cannot be performed; the number of co-occurrence words in all new requirement sentences that can perform semantic embedding and the emotion visualized writing thinking is counted respectively, and a ratio calculation is performed to obtain a semantic overlap degree; if the semantic overlap degree is higher than a preset overlap threshold and the subject is consistent, the corresponding new requirement sentence is marked as a structure embedding candidate sentence; the type of the connecting word of the structure embedding candidate sentence is judged, if it contains a parallel connecting word, a parallel insertion is set, if it contains a transition or causal connecting word, a subordinate insertion is set; the insertion position in the emotion visualized writing thinking is matched based on the type of the connecting word, the content order of the emotion visualized writing thinking after the insertion of the structure embedding candidate sentence is adjusted based on the language logic, and a complete writing thinking is obtained.

[0033] Further, the feedback evaluation manner comprises:

[0034] The complete writing thinking is matched with a writing logic template corresponding to the user's grade identifier, and a thinking score of the complete writing thinking is output as an evaluation result.

[0035] An AI interaction system for instant generation of a writing thinking, which is used to implement an AI interaction method for instant generation of a writing thinking, and characterized in that it comprises:

[0036] A data collection module, configured to collect writing task related input data of a user and perform data cleaning to obtain optimized writing task input data;

[0037] A semantic analysis module, configured to perform structured semantic analysis on the optimized writing task input data to obtain a writing task intention data set;

[0038] A user adaptation module, configured to obtain personal data of the user, perform cognitive degree matching on the writing task intention data set based on the personal data of the user, and output a user-adapted writing thinking;

[0039] An emotion promoting module, configured to promote the emotion rhythm of the user-adapted writing thinking to generate an emotion visualized writing thinking;

[0040] A semantic embedding module, configured to collect new writing requirements of the user in real time, perform semantic embedding on the emotion visualized writing thinking and the new writing requirements of the user, and obtain a complete writing thinking;

[0041] A feedback evaluation module, configured to perform feedback evaluation on the complete writing thinking, and send the evaluation result and the complete writing thinking to a preset user terminal; the modules are connected through wired and / or wireless modes.

[0042] The technical effects and advantages of the AI ​​interactive method and system for real-time generation of essay writing ideas in this invention are as follows:

[0043] By collecting user-input writing task-related data, this method performs targeted semantic analysis on the input data for each modality. Then, through cognitive adaptation and emotional rhythm progression, it provides reasonable essay writing strategies. Simultaneously, it acquires new writing requirements from users in real time and adjusts the original essay writing strategies to obtain the final complete essay writing approach. This achieves an AI-related adaptive generation method for essay writing strategies. Compared with existing experience, by introducing a combination of semantic analysis and image recognition, it more accurately identifies the text content in images and links various semantically relevant elements based on semantic relationships, resulting in a logically clear essay logic path diagram. By adding weight evaluation for each object in the image, it comprehensively considers the position, boundary contour complexity, and center distance of each object to determine its semantic priority, avoiding the problem of various semantic objects in the image being treated as equal in importance. This method addresses the issue of misaligned writing ideas. It employs more precise semantic segmentation of voice input data, significantly improving the output rate of usable writing ideas under voice input conditions. Simultaneously, it constructs a user cognitive ability vector and matches it with the essay intent complexity vector to generate essay ideas that better suit the user's current cognitive level, ensuring comprehension and application. Furthermore, it visualizes the abstract emotional changes within the generated essay ideas, allowing users to perceive the emotional progression within the writing process and indirectly enhancing their understanding of emotional shifts. A reasonable insertion mechanism for new user-added essay requirements is added, effectively avoiding semantic disconnect or logical inconsistencies caused by the forced patchwork of new content. In summary, this essay writing idea generation method significantly improves the efficiency and quality of writing idea generation, exhibiting higher semantic parsing accuracy, stronger personalized adaptation capabilities, and superior teaching guidance. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of an AI interactive method for real-time generation of essay writing ideas according to the present invention.

[0045] Figure 2 This is a schematic diagram of an AI interactive system for instantly generating essay writing ideas according to the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Example 1

[0048] Please see Figure 1 As shown in this embodiment, an AI-interactive method for real-time generation of essay writing ideas includes:

[0049] S1. Collect and clean the user's writing task-related input data to obtain optimized writing task input data;

[0050] S2. Perform structured semantic analysis on the input data of the optimized writing task to obtain the writing task intent dataset;

[0051] S3. Obtain user personal data, match the cognitive level of the essay task intent dataset with the user personal data, and output the user-adapted essay writing ideas;

[0052] S4. Adapt the user's essay writing process to the emotional rhythm and generate an emotionally visualized essay writing process.

[0053] S5. Real-time collection of new writing requirements from users, semantic integration of emotionally visualized essay writing ideas with new writing requirements from users to obtain complete essay writing ideas;

[0054] S6. Provide feedback and evaluation on the complete essay writing process, and send the evaluation results and the complete essay writing process to the preset user terminal.

[0055] In this embodiment, the original input data related to the writing task includes image data of the essay description content, image data of the materials used for writing essays based on pictures, and voice data directly input by the user. Noise suppression is performed on the image data, and silent segments in the voice data are removed and filtered to obtain higher quality image data and voice data, respectively.

[0056] Methods for performing structured semantic analysis include:

[0057] The data types of the input data for the writing task are optimized to include image data of the writing requirements, image data of the writing materials, and user voice input data. The image data of the writing requirements and the image data of the writing materials are high-quality image data obtained through data cleaning, and the user voice input data is high-quality voice data obtained through data cleaning.

[0058] The image data of the essay requirements is subjected to grayscale normalization to obtain character image units. Grayscale normalization improves the edge contrast of the image data of the essay requirements, making the characters clearer. The image of each character region is used as a character image unit.

[0059] Contour fitting is performed on character image units to generate character codes. The character codes are then arranged into character code segments. In this embodiment, an OCR model combined with deep learning is used to extract contour features for each character image unit. The contour features include font shape features such as stroke direction and corner points. The contour features are matched with a dictionary stored in the database to obtain the font shape of each character image unit, which is the character code. The character codes are then used as the required arrangement order of each character image unit in the image data to form character code segments.

[0060] The number of pixel blocks at the corresponding position of each character in the character encoding segment is detected and the pixel density at the corresponding position is calculated. If the pixel density is less than a preset readable density threshold, a low readability label is added to the character. The pixel density is obtained by counting the number of pixel blocks at the corresponding position of the character and calculating the ratio of the number of pixel blocks to the area of ​​the rectangular region surrounding the character. If the pixel density is less than the preset readable density threshold set based on known font recognition standards, it indicates that the character may have misrecognition or broken strokes, resulting in low readability. Therefore, a low readability label is added to the character. Conversely, if the pixel density is greater than or equal to the preset readable density threshold, it indicates that the font is clear and can be completely matched with the dictionary to determine the specific character of the corresponding font shape. Therefore, no processing is required.

[0061] The process involves identifying the glyph components of characters corresponding to low-readability tags and constructing candidate characters. These candidate characters are then combined with adjacent characters to form candidate word vectors. Specifically, the morphological distribution features of characters corresponding to low-readability tags are extracted, and basic stroke units such as horizontal, vertical, left-falling, right-falling, and dot strokes are used as glyph components. Based on a dictionary stored in the database, these glyph components are combined to obtain several candidate characters. To further refine the selection of candidate characters, adjacent characters at the position of each candidate character are used to construct character tuples with any one of the candidate characters. These character tuples are then converted into vector forms, which are the candidate word vectors.

[0062] The semantic similarity between the candidate word vector and the adjacent content word vectors is calculated. The candidate word vector with the highest semantic similarity is selected as the appropriate word vector and covers the original candidate word vector. The adjacent content word vector refers to the complete recognized word vector within the adjacent preset window area where the candidate word vector is located. The size of the preset window area is set based on the historical recognition experience of those skilled in the art. The cosine similarity between the candidate word vector and the adjacent content word vector is calculated. This cosine similarity is the semantic similarity at this time. By continuously changing the candidate characters in the candidate word vector, and using different candidate word vectors to calculate the semantic similarity each time, the candidate word vector with the highest semantic similarity is finally selected as the appropriate word vector and used to cover the original candidate word vector. At the same time, the candidate characters corresponding to the appropriate word vector are used as the determined characters to cover the characters with low-readability labels.

[0063] The subject, predicate, and object words of each clause in the adjusted character encoding segment are identified and a clause node chain is constructed. The adjusted character encoding segment is obtained after the candidate characters are confirmed. The adjusted character encoding segment is divided into several clauses based on the period. The subject, predicate, and object in each clause are identified. A ternary expression of the clause is established based on the subject, predicate, and object words in the clause. The subject, predicate, and object are all nodes, which constitute the clause node chain of the clause.

[0064] The semantic relationships between clauses are identified, and the clause node chains are connected based on these semantic relationships to generate the essay's logical path graph. Semantic relationships are determined by identifying the connecting words between adjacent clauses. For example, "because" and "therefore" indicate a causal relationship. Other relationships include progressive, concessive, and adversative relationships. If there is a semantic relationship between two clauses, a directed edge is constructed based on this semantic relationship, and all clause node chains with semantic relationships are connected to obtain the essay's logical path graph.

[0065] The image data of the writing materials is weighted and labeled to obtain the label-level thinking data; the intent structure of the user's voice input data is extracted to obtain the structure-formed thinking data; based on the data type, any one of the following—the writing logic path diagram, the label-level thinking data, and the structure-formed thinking data—is selected as the writing task intent dataset. In this embodiment, the three types of data are not processed all at once, but rather three processing branches are constructed. By identifying and optimizing the data type of the writing task input data in the early stage of structured semantic analysis, the corresponding processing branch is selected for processing to obtain the writing intent dataset of this data type, which improves the flexibility of performing structured semantic analysis.

[0066] Methods for adjusting tag weights include:

[0067] In this embodiment, an edge detection algorithm is used to extract continuous closed shape regions with significant edge changes in the image data of the writing material, and to identify whether the region has semantics. For example, people, animals, flowers and certain objects all have clear semantic information. It is determined whether these continuous closed shape regions have such semantic information, and the continuous closed region is regarded as a semantic object. Then, the set of contour points of the semantic object is extracted, which is the edge contour.

[0068] Pixel segmentation is performed based on the edge contours of semantic objects to generate image mask regions for each semantic object. Semantic labels are added to the corresponding image mask regions. In this process, a deep learning-based image semantic segmentation model (such as the DeepLabv3+ model) is used to perform pixel-level classification processing on each semantic object region based on the edge contours of the semantic objects. The original essay material image data is divided into several image mask regions, and the semantic labels of the semantic objects are added to the corresponding image mask regions.

[0069] The ratio of each image mask region to the area of ​​the essay material image data is calculated to obtain the pixel region proportion. The pixel region proportion is based on the ratio of the image mask region of each semantic object to the area of ​​the essay material image data, which is mapped to the main visual weight in the essay idea. The larger the pixel region proportion, the more likely the corresponding semantic object is to become the writing clue that the user chooses first.

[0070] A boundary contour complexity function is constructed based on the area of ​​the image mask region, and this function is used to calculate the boundary contour complexity of the image mask region. The formula for calculating the boundary contour complexity function is as follows: ;in, Represents the boundary contour complexity of any image mask region; This represents the perimeter of the boundary of the aforementioned image mask region; This represents the area of ​​the image mask region mentioned above. It should be noted that the above formula uses dimensionless calculation. The larger the boundary contour complexity value, the more concave and convex structures, more obvious posture changes, or more complex curved shapes the semantic object boundary of the corresponding image mask region has. Such semantic objects often belong to the form of humans or animals, including multiple limbs with branches, and may often be the focus of writing.

[0071] The center distance is obtained by calculating the distance between the center coordinates of each image mask region and the center coordinates of the essay material image data. The center coordinates of the image mask region refer to the geometric centroid coordinates of the image mask region. The center distance is obtained by calculating the Euclidean distance between the geometric centroid coordinates and the coordinates of the center point of the essay material image data. The center distance is used to determine the degree of offset of the semantic object relative to the center of the image. The smaller the center distance, the closer the corresponding semantic object is to the center. Visually, objects that are closer to the center have a higher attentional attraction for users.

[0072] The comprehensive semantic weight of the corresponding image mask region is obtained by weighted fusion of pixel region proportion, boundary contour complexity and center distance. The weight required for weighted fusion is obtained by calculating the average value of multiple historical weighted records. The comprehensive semantic weight obtained by weighted fusion is used to represent the probability of the corresponding image mask region being used by students as a writing clue.

[0073] All image mask regions are sorted based on comprehensive semantic weights to obtain a semantic priority sequence. The semantic priority sequence obtained by sorting all image mask regions in descending order based on comprehensive semantic weights forms a picture-based writing guidance flow to guide students to obtain the correct writing ideas.

[0074] The semantic priority sequence is matched with the preset image writing template to obtain the tag-level thinking data. The preset image writing template refers to a dataset of writing elements for image writing stored in the database, including information such as writing theme, writing object, extended meaning of elements in the image, and historical image writing model essays. The semantic priority sequence here is matched with the preset image writing template. Based on the order of each semantic object and semantic label in the semantic priority sequence, an outline of the writing process that conforms to a certain theme is constructed, which is the tag-level thinking data.

[0075] Methods for extracting intent structure include:

[0076] The user's voice input data is converted into voice text content. The pause interval is detected and the voice text content is segmented into sentences based on the pause interval to obtain the voice sentence content. In this embodiment, the user's voice input data is converted into voice text content through an acoustic model, and the silence duration and pause boundaries in the voice stream are detected to obtain the sentence segmentation signal. The sentence segmentation is performed based on the sentence segmentation signal to obtain the voice sentence content of each sentence.

[0077] The speech sentence content is segmented into words, and each word is converted into a word vector. The word vectors belonging to the same speech sentence content are combined into a multi-dimensional semantic representation vector. The Chinese word segmentation model based on BiLSTM divides the speech sentence content of each sentence into several words, and maps each word into a word vector. Then, the word vectors belonging to the same speech sentence content are concatenated into a multi-dimensional semantic representation vector with a dimension equal to the number of word vectors.

[0078] The semantic distance between any two multidimensional semantic representation vectors is calculated. Multidimensional semantic representation vectors whose semantic distance to any one of them is less than a preset semantic distance threshold are integrated into a high-density vector set. The speech region corresponding to the high-density vector set is then identified. Specifically, the semantic distance is obtained by calculating the Euclidean distance between any two multidimensional semantic vectors. A range scan is performed on any multidimensional semantic representation vector to identify other multidimensional semantic vectors whose semantic distance to the given multidimensional semantic vector is less than a preset semantic distance threshold set based on historical experience. These multidimensional semantic vectors are then grouped into a high-density vector set. Multiple high-density vector sets are obtained through semantic distance filtering. Each high-density set is then mapped to the original speech text content to confirm the speech region corresponding to each high-density set.

[0079] The semantic center vector of each high-density vector set is identified and a semantic center sequence is constructed. The semantic center vector is obtained by calculating the average vector of all multidimensional semantic representation vectors in each high-density vector set. The semantic center sequence is obtained by sorting the speech regions corresponding to the high-density vector sets in the complete speech text content. This semantic center sequence reflects the hierarchical changes in the speech text content.

[0080] The semantic change direction is determined based on the angle change of adjacent semantic center vectors in the semantic center sequence, and a semantic center change path is constructed. The position where the angle change is higher than the preset angle threshold in the semantic center change path is identified and the position is determined as the structural turning point. The semantic center vectors in the semantic center sequence are connected to obtain the semantic center change path, and the angle change between any two adjacent semantic center vectors is calculated. If the angle change is higher than the preset angle threshold set based on historical experience, it indicates that the logic or core writing intention has changed significantly. Therefore, the position of the specific text content corresponding to the two semantic center vectors with significant angle changes is taken as the structural turning point.

[0081] The segment regions corresponding to the high-density vector sets before and after the structural transition points are reordered according to linguistic logic to obtain structural formation thought data. The logic of the segment regions corresponding to the high-density vector sets before and after the structural transition points is re-examined. The logical rationality of the above segment regions is judged according to existing known linguistic logic standards. If it is reasonable, no processing is done. If it is unreasonable, it is reordered based on known linguistic logic standards to ensure that the logic conforms to the linguistic logic standards. For example, when a user inputs with voice, if they forget a part of the content at the beginning and then fill it in later, the logic will be disordered. In this case, it is necessary to reorder these contents according to the logic of the normal development sequence of events.

[0082] Methods for matching cognitive levels include:

[0083] User personal data includes historical essay texts, historical essay scores, user grade level identifiers, and average writing time. The system calculates the number of sentences, average sentence length, and number of sentence structure types for any given historical essay text. It also constructs a complex sentence structure parameter set, using periods as identifiers to divide the historical essay text into sentences, counting the number of sentences, detecting the length of each sentence, and calculating the average sentence length. Based on existing language knowledge, it identifies the sentence structure types of these sentences, such as coordinate sentences, nested sentences, and relative clauses.

[0084] The ratio of functional words to the total number of words in a historical essay text is used to obtain the limited word density. Functional words refer to modal adverbs such as "very" or "some," logical connectors such as "because" or "however," and time adverbs that express time. These words do not carry semantic content but reflect the user's expressive tendencies in text organization and emotional control. The limited word density, obtained by calculating the ratio of functional words to the total number of words, is used to measure the user's ability to utilize writing skills.

[0085] Identify the frequency of semantic relationship texts in historical essays. The frequency of semantic relationship texts refers to the number of times that indicative words of semantic relationships in existing language knowledge, such as causal relationships, adversative relationships, or progressive relationships, appear.

[0086] The sentence complexity parameter set, limited word density, and frequency of occurrence are normalized to obtain a language structure parameter set. These parameters are then summed and averaged to obtain the user's average language structure parameter set. Specifically, by normalizing the sentence complexity parameter set, limited word density, and frequency of occurrence, a unified dimensional language structure parameter set is obtained. Furthermore, the parameters of corresponding dimensions in all language structure parameter sets are summed and averaged to obtain a user's average language structure parameter set. This parameter set is used to reflect the user's comprehensive evaluation level in text structure processing ability. It should be noted that the acquisition of user personal data here is subject to a disclaimer: obtaining user authorization or consent to the use of user personal data is required.

[0087] The standard deviation of essay scores is calculated based on historical essay scores, and the ratio of the average essay score to the highest essay score is calculated to obtain the score fluctuation deviation. The standard deviation of essay scores is used to represent the stability of the user's writing level, and the score fluctuation deviation reflects the gap between the user's stable level and the upper limit of writing ability. It should be noted that the historical essay score refers to the score of the historical essay text completed by the user in the history record in this embodiment. This score is obtained by scoring based on the existing essay teaching scoring standards in the teaching field.

[0088] Using the user's grade level as an index, the absolute value of the difference between the current user's average writing time and the standard writing time of students in the same grade is calculated to obtain the time deviation. The standard writing time set for students in the same grade is queried in the database using the user's grade level as an identifier, and the time deviation is calculated based on the current user's average writing time and the standard writing time of students in the same grade. The larger the value, the greater the gap between the user's thinking speed in conceiving and organizing ideas and the standard requirements.

[0089] A user cognitive ability vector is constructed based on the user's average language structure parameter set, score standard deviation, score fluctuation deviation, and duration deviation, and the user's grade identifier is used as the vector label. The user's average language structure parameter set, score standard deviation, score fluctuation deviation, and duration deviation are concatenated to form the user cognitive ability vector, which is used to represent the current user's cognitive ability strength in writing.

[0090] This study extracts the writing element structure, plot progression steps, and number of logical connection types from the essay task intent dataset and constructs an essay intent complexity vector. The writing element structure includes the frequency of occurrence of elements that play a crucial role in the essay structure, such as participating characters, key items, and writing scenarios. The plot progression steps refer to the number of events required to complete the entire essay. The number of logical connection types refers to the frequency of semantic relationships such as causal or progressive relationships. These three dimensions are combined to form the essay intent complexity vector, reflecting the complexity of the current essay task. A cognitive bias index is output by comparing the user's cognitive ability vector with the essay intent complexity vector. Based on this cognitive bias index, the essay task intent dataset is adjusted to output a user-adapted essay writing approach.

[0091] Methods for outputting cognitive bias indicators include:

[0092] The cognitive ability vector of the user is evaluated to output the cognitive ability level. The cognitive ability level corresponding to the user's cognitive ability vector is determined by mapping the parameter values ​​in the user's cognitive ability vector to a cognitive ability evaluation parameter table set based on historical experience. The cognitive ability level is a constant.

[0093] The complexity of the essay intent complexity vector is estimated, and the task structure level is output. Based on historical evaluation experience, corresponding weights are added to each dimension parameter in the essay intent complexity vector. The task structure level is obtained by summing all weighted dimension parameters.

[0094] The difference between the cognitive ability level and the task structure level is the cognitive bias index, where both the cognitive ability level and the task structure level are normalized and reduced to a minimum. The difference between the two levels is then calculated, which gives the cognitive bias index. A higher and positive value indicates that the cognitive ability is far higher than the requirements of the current writing task. A negative value indicates that the cognitive ability is lower than the requirements of the current writing task.

[0095] Methods for adjusting intent include:

[0096] If the cognitive bias index is greater than zero, no processing is performed; otherwise, the writing task intent dataset is simplified to obtain a user-adapted writing approach. Simplification refers to adjusting the writing task intent dataset while adhering to known linguistic logic, such as reducing the number of writing elements, merging scene fragments, or replacing complex words. The adjusted writing task intent dataset is then determined to be the user-adapted writing approach.

[0097] Methods for advancing the emotional rhythm include:

[0098] The user-adapted essay writing ideas are divided into idea content units, and the emotional descriptions in each idea content unit are identified. Since the data sources of user-adapted essay writing ideas may be essay logic path diagrams, tag-level idea data, or structured idea data, the method of dividing idea content units is different for each type of user-adapted essay writing ideas. If the user-adapted essay writing ideas originate from the essay logic path diagram, the semantic relationships that are directed edges in the essay logic path diagram are used as the basis for division into several idea content units.

[0099] If the user's essay writing approach originates from tag-based hierarchical thinking data, the comprehensive semantic weight value corresponding to the tag-based hierarchical thinking data is divided into several segments according to historical segmentation experience. The tag-based hierarchical thinking data corresponding to all comprehensive semantic weights in each segment constitutes a thinking content unit. If the user's essay writing approach originates from structured thinking data, it is divided according to each high-density vector set. The paragraph region corresponding to each high-density vector set constitutes a thinking content unit. The system checks whether each thinking content unit contains words that can express emotional color, such as emotional adjectives, emotional adverbs, or attitude words. If so, the content of the corresponding thinking content unit is treated as emotional description content.

[0100] The system detects the emotion type of emotional descriptions and outputs the emotion intensity by matching them with a preset emotion level table. The preset emotion level table is an emotion assessment template based on a large number of essays stored in the database. The system uses this preset emotion level table to match emotional words or phrases and sentences containing emotional words in the essays and outputs the corresponding emotion type and emotion intensity. For example, the emotion intensity for "happy" is 4, while the emotion intensity for "very happy" is 5, and this emotion type is classified as joy.

[0101] Using thought content units as nodes and emotion type as labels, an emotion intensity fluctuation curve is plotted based on the structural order of thought content units in the user-adapted essay writing thought process. In the emotion intensity fluctuation curve, the X-axis represents the index number of the thought content unit, the Y-axis represents the emotion intensity, and an emotion type label is added to each node.

[0102] The slope of the emotion intensity fluctuation curve is calculated, and nodes with a slope higher than a preset slope threshold are marked as emotion mutation nodes. The slope is obtained by calculating the ratio of the change in emotion intensity between each adjacent node in the emotion intensity fluctuation curve to the difference between the index numbers of the adjacent nodes. The preset slope threshold is set based on historical emotion assessment experience. If the slope is higher than the preset slope threshold, it indicates that the emotion intensity changes drastically, and such nodes are marked as emotion mutation nodes.

[0103] The node with the highest emotional intensity in the emotional intensity fluctuation curve is identified as the emotional climax node. This is achieved by traversing all nodes in the emotional intensity fluctuation curve and determining the node with the highest emotional intensity as the emotional climax node. If multiple nodes have the same maximum emotional intensity, the node with the later index number is determined as the emotional climax node.

[0104] Among all emotional abrupt change nodes in the curve segment following the emotional climax node, the node whose change in emotional intensity with the previous node is higher than a preset intensity change threshold and whose slope is negative is identified as an emotional turning point. Specifically, by traversing the remaining segment after the emotional climax node in the emotional intensity fluctuation curve, all emotional abrupt change nodes are identified, and the emotional intensity of the node is compared with the emotional intensity of the previous node. If the change in emotional intensity between the two nodes is higher than a preset intensity change threshold set based on historical experience, and the slope of the emotional abrupt change node is negative (i.e., downward), then the emotional abrupt change node is determined as an emotional turning point.

[0105] By mapping emotional mutation nodes, emotional climax nodes, emotional turning points, and emotional types back to the thought content units and adding corresponding indexes, we obtain the emotional visualization essay writing thought process. In this process, emotional mutation nodes, emotional climax nodes, emotional turning points, and emotional types are used as index labels and marked on the corresponding thought content units to obtain the labeled user-adapted essay writing thought process, which is the emotional visualization essay writing thought process.

[0106] Methods for semantic embedding include:

[0107] The user's new writing requirements are converted into text content, and then segmented into sentences based on grammatical pauses to obtain new requirement sentences. In this embodiment, the user's new writing requirements specifically refer to voice type data, because it is more reasonable to add writing requirements by having the user input voice type data in practice. First, the user's new writing requirements are converted into text content, and then the text content is segmented into sentences based on the pauses in the original voice type data to obtain several new requirement sentences.

[0108] Extract the subject from the newly added requirement clause and determine whether the subject is consistent with the subject of the emotional visualization essay writing approach. If they are inconsistent, semantic embedding cannot be performed. Since the emotional visualization essay writing approach may include multiple subjects, the method here is as follows: First, identify the subject of the newly added requirement clause. If the subject does not exist in the emotional visualization essay writing approach and also does not exist in subsequent newly added requirement clauses, then the newly added requirement clause is considered inconsistent with the subject of the emotional visualization essay writing approach, and semantic embedding cannot continue. If the subject of the newly added requirement clause already exists in the emotional visualization essay writing approach or only exists in subsequent newly added requirement clauses, then the subject of the newly added requirement clause is considered consistent with the subject of the emotional visualization essay writing approach, and semantic embedding can continue.

[0109] The number of co-occurring words in all newly added requirement clauses that can be semantically embedded and in the emotional visualization essay writing ideas are counted separately, and the ratio is calculated to obtain the semantic overlap degree. In this embodiment, words or synonyms that appear together in the newly added requirement clauses that can be semantically embedded and in the emotional visualization essay writing ideas are identified as co-occurring words. The number of co-occurring words is counted, and the ratio of the number of co-occurring words to the total number of words in the emotional visualization essay writing ideas is calculated, which is the semantic overlap degree.

[0110] If the semantic overlap is higher than the preset overlap threshold and the subject is consistent, the corresponding newly added requirement clause is marked as a candidate clause for structural embedding. The preset overlap threshold is set based on historical semantic embedding experience. If the semantic overlap is higher than the preset overlap threshold and the subject is consistent, it means that the corresponding newly added requirement clause can be semantically embedded normally, and the newly added requirement clause is judged as a candidate clause for structural embedding.

[0111] The method determines the type of conjunction in the candidate clauses for structural embedding. If a parallel conjunction is included, parallel insertion is set; if a contrastive or causal conjunction is included, subordinate insertion is set. The conjunction types include logical conjunctions commonly used in existing language logic, such as parallel conjunctions, contrastive conjunctions, and causal conjunctions. In this embodiment, parallel conjunctions are set as parallel insertions. Parallel insertion means that for sentences with similar semantic content or identical behaviors, the candidate clauses for structural embedding can be inserted into adjacent positions of the original sentence to form a semantic structure of parallel description. Contrastive or causal conjunctions are set as subordinate insertions. Subordinate insertion means that for complex structures such as contrastive or causal, the candidate clauses for structural embedding need to be inserted into positions that can form an effective main-subordinate structure with the original sentence. By explicitly specifying the insertion method, grammatical connection errors are avoided.

[0112] Based on the insertion position in the emotional visualization essay writing process using conjunction type matching, and by adjusting the content order of the inserted structure after embedding candidate clauses based on linguistic logic, a complete essay writing process is obtained. Specifically, by determining the insertion position in the emotional visualization essay writing process using conjunction type selection, and based on existing linguistic logic knowledge, the content order is adjusted to ensure logical coherence and eliminate grammatical problems, resulting in the final complete essay writing process.

[0113] Methods for conducting feedback evaluation include:

[0114] The complete essay writing approach is matched with the essay logic template corresponding to the user's grade level. The resulting score is the essay logic template, which is the evaluation result. The essay logic template refers to the standard essay writing template for each grade level stored in the database. This template includes standard writing information such as writing structure and format, usage of conjunctions, and rules for the rhythm of emotional changes. The template also includes a scoring range for each dimension of the standard writing information. By querying the essay logic template for the user's grade level, the complete essay writing approach is matched with that template. The scores for each dimension are summed to obtain the essay logic template score, which is the evaluation result. This evaluation result is used to provide feedback to the user and generate targeted suggestions for improvement, indirectly enhancing the user's writing skills.

[0115] This embodiment collects writing task-related data input by users, performs targeted semantic analysis on the input data for each modality, and then provides reasonable essay writing ideas through cognitive adaptation and emotional rhythm progression. Simultaneously, it acquires new writing requirements from the user in real time and adjusts the original essay writing ideas to obtain the final complete essay writing approach, realizing an AI-related adaptive generation method for essay writing ideas. Compared with existing experience, by introducing a method combining semantic analysis and image recognition, it more accurately identifies the text content in images and links various semantically related elements based on semantic relationships, obtaining a logically clear essay logic path diagram. By adding weight evaluation for each object in the image, the semantic priority of each object is obtained by comprehensively considering its position, boundary contour complexity, and center distance, avoiding the confusion of different semantic elements in the image without prioritizing them. This method addresses the issue of misaligned writing ideas. It employs more precise semantic segmentation of voice input data, significantly improving the output rate of usable writing ideas under voice input conditions. Simultaneously, it constructs a user cognitive ability vector and matches it with the essay intent complexity vector to generate essay ideas that better suit the user's current cognitive level, ensuring comprehension and application. Furthermore, it visualizes the abstract emotional changes within the generated essay ideas, allowing users to perceive the emotional progression within the writing process and indirectly enhancing their understanding of emotional shifts. A reasonable insertion mechanism for newly added essay requirements is added, effectively avoiding semantic disconnect or logical inconsistencies caused by the forced patchwork of new content. In conclusion, this essay writing idea generation method significantly improves the efficiency and quality of writing idea generation, exhibiting higher semantic parsing accuracy, stronger personalized adaptation capabilities, and superior teaching guidance.

[0116] Example 2

[0117] Please see Figure 2 As shown, for parts not described in detail in this embodiment, please refer to the description in Embodiment 1. This embodiment provides an AI interactive system for real-time generation of essay writing ideas, including:

[0118] The data acquisition module is used to collect input data related to the user's writing task and perform data cleaning to obtain optimized input data for the writing task.

[0119] The semantic analysis module is used to perform structured semantic analysis on the input data of the optimized writing task to obtain the writing task intent dataset;

[0120] The user adaptation module is used to acquire users’ personal data, match the cognitive level of the essay task intent dataset with the user’s personal data, and output user-adapted essay writing ideas.

[0121] The Emotional Flow Module is used to advance the emotional rhythm of the essay writing process to match the user's writing ideas, and generate an emotionally visualized essay writing process.

[0122] The semantic embedding module is used to collect new writing requirements from users in real time, and semantically embed the emotional visualization essay writing ideas with the new writing requirements from users to obtain a complete essay writing ideas.

[0123] The feedback and evaluation module is used to provide feedback and evaluation on the complete essay writing process, and sends the evaluation results and the complete essay writing process to the preset user terminal; the modules are connected to each other via wired and / or wireless means.

[0124] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0125] 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, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0126] In the description of this invention, it should be understood that the terms "first," "second," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance.

[0127] In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0128] In the description of this invention, "several" means one or more, and "a large number" means two or more.

[0129] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0130] All formulas in this manual are dimensionless and calculated numerically. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0131] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. An AI-interactive method for instant generation of essay writing ideas, characterized in that, include: S1. Collect and clean the user's writing task-related input data to obtain optimized writing task input data; S2. Perform structured semantic analysis on the input data of the optimized writing task to obtain the writing task intent dataset; S3. Obtain user personal data and perform cognitive level matching on the essay task intent dataset based on the user personal data, including: User personal data includes historical essay texts, historical essay scores, user grade level identifiers, and average writing time; the number of sentences, average sentence length, and number of sentence structure types in any historical essay text are statistically analyzed, and a complex sentence structure parameter set is constructed; the ratio of the number of functional words to the total number of words in historical essay texts is calculated to obtain the limited word density; the frequency of occurrence of semantic relationship texts in historical essay texts is identified; the complex sentence structure parameter set, limited word density, and frequency of occurrence are normalized to obtain a language structure parameter set, which is then summed and averaged to obtain the user's average language structure parameter set; The standard deviation of essay scores is calculated based on historical essay scores, and the ratio of the average essay score to the highest essay score is calculated to obtain the score fluctuation deviation. Using the user's grade identifier as an index, the absolute value of the difference between the current user's average writing time and the standard writing time of students in the same grade is calculated to obtain the time deviation. Based on the user's average language structure parameter set, score standard deviation, score fluctuation deviation, and time deviation, a user cognitive ability vector is constructed and the user's grade identifier is used as the vector label. Extract the writing element structure, plot progression steps, and number of logical connection types from the writing task intent dataset and construct the writing intent complexity vector; compare the user cognitive ability vector with the writing intent complexity vector to output a cognitive bias index; adjust the writing task intent dataset based on the cognitive bias index to output a user-adapted writing approach. S4. Adapt the writing process to the user's emotional rhythm, including: The user-adapted essay writing process is divided into content units, and the emotional descriptions within each content unit are identified. The emotional type of these descriptions is detected and its intensity is output by matching them with a preset emotional level table. Content units are used as nodes and labeled with their emotional types. An emotional intensity fluctuation curve is plotted based on the structural order of these content units within the user-adapted essay writing process. The slope of the emotional intensity fluctuation curve is calculated, and nodes with slopes exceeding a preset threshold are marked as emotional abrupt change nodes. The node with the highest emotional intensity in the emotional intensity fluctuation curve is identified as the emotional climax node. Among all emotional abrupt change nodes in the curve segment following the emotional climax node, nodes whose change in emotional intensity from the previous node exceeds a preset threshold and whose slope is negative are identified as emotional turning points. Emotional abrupt change nodes, emotional climax nodes, emotional turning points, and emotional types are mapped back to content units and corresponding indexes are added to obtain an emotionally visualized essay writing process. S5. Real-time collection of new writing requirements from users, semantic integration of emotionally visualized essay writing ideas with new writing requirements from users to obtain complete essay writing ideas; S6. Provide feedback and evaluation on the complete essay writing process, and send the evaluation results and the complete essay writing process to the preset user terminal.

2. The AI ​​interactive method for real-time generation of essay writing ideas according to claim 1, characterized in that, The methods for performing structured semantic analysis include: Optimize the data types of input data for writing tasks, including image data of essay requirements, image data of essay materials, and user voice input data; The image data for the essay requirements is grayscale normalized to obtain character image units. Contour fitting is performed on the character image units to generate character codes, which are then arranged into character code segments. The number of pixel blocks at each position corresponding to a character in the character code segment is detected, and the pixel density at that position is calculated. If the pixel density is less than a preset readable density threshold, a low-readability label is added to that character. The glyph components of the characters corresponding to the low-readability labels are identified, and candidate characters are constructed. Candidate characters are combined with adjacent characters at their corresponding positions to form candidate word vectors. The semantic similarity between the candidate word vector and the word vectors of adjacent content is calculated. The candidate word vector with the highest semantic similarity is selected as the appropriate word vector and overwrites the original candidate word vector. The subject, predicate, and object words of each clause in the adjusted character code segment are identified, and a clause node chain is constructed. The semantic relationships between clauses are identified, and the clause node chains are connected based on these semantic relationships to generate an essay logic path diagram. The image data of the writing materials are labeled and weighted to obtain the label-level thinking data; the intent structure of the user's voice input data is extracted to obtain the structure-formed thinking data; based on the data type, any one of the following can be selected as the writing task intent dataset: the writing logic path diagram, the label-level thinking data, and the structure-formed thinking data.

3. The AI ​​interactive method for real-time generation of essay writing ideas according to claim 2, characterized in that, The methods for adjusting label weights include: This process involves identifying semantic objects in the essay writing material image data and extracting the edge contours of each object. Based on these edge contours, pixel segmentation is performed to generate image mask regions for each semantic object, and semantic labels are added to the corresponding image mask regions. The ratio of the area of ​​each image mask region to the area of ​​the essay writing material image data is calculated to obtain the pixel region proportion. A boundary contour complexity function is constructed based on the area of ​​the image mask regions, and this function is used to calculate the boundary contour complexity of the image mask regions. The distance between the center coordinates of each image mask region and the center coordinates of the essay writing material image data is calculated to obtain the center distance. A weighted fusion of the pixel region proportion, boundary contour complexity, and center distance is performed to obtain the comprehensive semantic weight of the corresponding image mask region. All image mask regions are sorted based on the comprehensive semantic weight to obtain a semantic priority sequence. The semantic priority sequence is matched with a preset image writing template to obtain label-based hierarchical thinking data.

4. The AI ​​interactive method for real-time generation of essay writing ideas according to claim 3, characterized in that, The methods for extracting intent structure include: The user's voice input data is converted into voice text content. Pause intervals are detected and the voice text content is segmented into sentences based on the pause intervals to obtain voice sentence content. The voice sentence content is then processed by word segmentation, and each word is converted into a word vector. The word vectors belonging to the same voice sentence content are combined into a multi-dimensional semantic representation vector. Calculate the semantic distance between any two multidimensional semantic representation vectors. Integrate multidimensional semantic representation vectors whose semantic distance to any other multidimensional semantic representation vector is less than a preset semantic distance threshold into a high-density vector set, and identify the segment regions corresponding to the high-density vector set. Identify the semantic center vector of each high-density vector set and construct a semantic center sequence. Determine the direction of semantic change based on the angle change of adjacent semantic center vectors in the semantic center sequence, and construct a semantic center change path. Identify positions in the semantic center change path where the angle change is higher than a preset angle threshold, and determine these positions as structural turning points. Reorder the segment regions corresponding to the high-density vector sets before and after the structural turning points according to linguistic logic to obtain structural formation idea data.

5. The AI ​​interactive method for real-time generation of essay writing ideas according to claim 4, characterized in that, The methods for outputting the cognitive bias index include: The cognitive ability vector of the user is evaluated to output the cognitive ability level; the complexity of the writing intention vector is estimated to output the task structure level; the difference between the cognitive ability level and the task structure level is calculated, which is the cognitive bias index. Methods for adjusting intent include: If the cognitive bias index is greater than zero, no processing is performed; otherwise, the thought process of the essay task intent dataset is simplified to obtain a user-adapted essay writing approach.

6. The AI ​​interactive method for real-time generation of essay writing ideas according to claim 5, characterized in that, The methods for semantic embedding include: The user's new writing requirements are converted into text content, and the sentences are divided based on grammatical pauses to obtain new requirement sentences. The subject of the new requirement sentence is extracted, and it is determined whether the subject is consistent with the subject of the emotional visualization essay writing idea. If they are inconsistent, semantic embedding is not possible. The number of co-occurring words in all new requirement sentences that can be semantically embedded and the emotional visualization essay writing idea is counted, and the ratio is calculated to obtain the semantic overlap. If the semantic overlap is higher than the preset overlap threshold and the subject is consistent, the corresponding new requirement sentence is marked as a candidate sentence for structural embedding. The type of conjunctions in the candidate sentence for structural embedding is determined. If it contains parallel conjunctions, it is set to parallel insertion. If it contains adversative or causal conjunctions, it is set to subordinate insertion. The insertion position in the emotional visualization essay writing idea is matched based on the conjunction type. The content order of the emotional visualization essay writing idea after inserting the candidate sentence for structural embedding is adjusted based on language logic to obtain the complete essay writing idea.

7. The AI ​​interactive method for real-time generation of essay writing ideas according to claim 6, characterized in that, The methods for conducting feedback evaluation include: The complete essay writing approach is matched with the essay logic template corresponding to the user's grade level, and the approach score of the complete essay writing approach is output, which is the evaluation result.

8. An AI interactive system for instant generation of essay writing ideas, used to implement the AI ​​interactive method for instant generation of essay writing ideas as described in any one of claims 1-7, characterized in that, include: The data acquisition module is used to collect input data related to the user's writing task and perform data cleaning to obtain optimized input data for the writing task. The semantic analysis module is used to perform structured semantic analysis on the input data of the optimized writing task to obtain the writing task intent dataset; The user adaptation module is used to acquire users’ personal data, match the cognitive level of the essay task intent dataset with the user’s personal data, and output user-adapted essay writing ideas. The Emotional Flow Module is used to advance the emotional rhythm of the essay writing process to match the user's writing ideas, and generate an emotionally visualized essay writing process. The semantic embedding module is used to collect new writing requirements from users in real time, and semantically embed the emotional visualization essay writing ideas with the new writing requirements from users to obtain a complete essay writing ideas. The feedback and evaluation module is used to provide feedback and evaluation on the complete essay writing process, and sends the evaluation results and the complete essay writing process to the preset user terminal; the modules are connected to each other via wired and / or wireless means.

Images