An electronic book content real-time simplification and enhancement method based on reading ability assessment

Abstract

This application discloses a method for real-time simplification and enhancement of e-book content based on reading ability assessment, relating to the field of educational technology. The method includes the following steps: Step one, real-time collection of multimodal behavioral data during the user's reading process. This multimodal behavioral data includes reading speed, page dwell time, repeated reading behavior targeting specific words or sentences, frequency and duration of new word clicks, and accuracy and reaction time of embedded comprehension tests. By sensing the user's reading behavior in real time, this application can dynamically assess their reading ability and accurately match it with text difficulty. This automatically simplifies the text to reduce cognitive load when the user encounters difficulties and enhances the content to stimulate deeper thinking when the user is proficient. This not only significantly improves reading fluency and comprehension depth but also achieves continuous adaptation of reading materials to the user's ability level through a closed-loop optimization system and personalized recommendations.

Description

Technical Field

[0001] This application relates to the field of educational technology, and in particular to a method for real-time simplification and enhancement of e-book content based on reading ability assessment. Background Technology

[0002] Traditional e-book readers primarily offer static content presentation, lacking the ability to perceive and adapt to the user's reading state. While some existing adaptive reading systems can manually adjust text difficulty according to preset reading levels, these adjustments are often static and one-size-fits-all, failing to dynamically intervene based on the user's real-time comprehension difficulties during the reading process. Furthermore, the assessment of user reading ability relies heavily on standardized post-reading tests rather than continuous analysis throughout the natural reading process. This results in an inaccurate and untimely match between reading materials and the reader's actual ability and immediate state, making it difficult to provide truly personalized adaptive reading support. Summary of the Invention

[0003] To address the aforementioned issues, this application provides a method for real-time simplification and enhancement of e-book content based on reading ability assessment.

[0004] This application provides a method for real-time simplification and enhancement of e-book content based on reading ability assessment, which adopts the following technical solution:

[0005] A method for real-time simplification and enhancement of e-book content based on reading ability assessment includes the following steps:

[0006] Step 1: Collect multimodal behavioral data of users during the reading process in real time. Multimodal behavioral data includes reading speed, page dwell time, repeated reading behavior for specific words or sentences, frequency and duration of new word clicks, accuracy and reaction time of embedded comprehension test;

[0007] Step 2: Based on the multimodal behavioral data collected in Step 1, a reading ability assessment model is used to dynamically construct and update the user's real-time reading ability profile. The reading ability profile is a multi-dimensional vector that includes vocabulary mastery, syntactic comprehension, information processing speed, and background knowledge familiarity.

[0008] Step 3: Perform real-time analysis on the currently presented e-book text content, and calculate the real-time difficulty feature value of the text through a text difficulty analysis model. The real-time difficulty feature value includes lexical complexity, sentence length and structural complexity, and semantic density.

[0009] Step 4: Calculate the matching degree between the real-time reading ability profile generated in Step 2 and the real-time difficulty feature value obtained in Step 3. When the matching degree is lower than the first preset threshold, generate a content simplification instruction; when the matching degree is higher than the second preset threshold, generate a content enhancement instruction.

[0010] Step 5: In response to the content simplification or content enhancement instructions generated in Step 4, call the corresponding AI content generation model to perform real-time conversion of the original text;

[0011] Step six: Collect feedback data from users reading the text transformed in step five, and use the feedback data to optimize the reading ability assessment model, text difficulty analysis model, and AI content generation model to form an adaptive closed-loop optimization system;

[0012] Step 7: Based on real-time reading ability profiles, actively select and recommend reading materials with matching difficulty from the e-book resource library to build personalized reading sequences. The recommendation system continuously optimizes its recommendation strategy based on user feedback on the recommended materials.

[0013] As a preferred technical solution of this application, step two specifically includes: the reading ability assessment model adopts a time series-based recurrent neural network model, takes the user's historical reading behavior data sequence as input, and outputs a multi-dimensional vector; the reading ability profile is dynamically updated at configurable time intervals to ensure that it reflects the user's latest reading ability status.

[0014] As a preferred technical solution of this application, step four further includes: using a cosine similarity algorithm to calculate the similarity between the reading ability profile vector and the real-time difficulty feature value vector; and setting the first preset threshold and the second preset threshold in a personalized manner according to the user's age group or reading ability baseline.

[0015] As a preferred technical solution of this application, in step five, the content simplification instructions include vocabulary simplification, syntax simplification, and insertion of explanations; vocabulary simplification specifically involves selecting simple words from a predefined thesaurus that match the user's vocabulary proficiency and automatically replacing complex words in the original text; syntax simplification specifically involves using natural language processing technology to automatically segment long and difficult sentences, simplify clause structures, or adjust word order to reduce sentence structure complexity; insertion of explanations specifically involves dynamically inserting floating window-style text explanations or illustrations next to complex concepts or words.

[0016] As a preferred technical solution of this application, in step five, the content enhancement instructions include detail enhancement, background expansion, and in-depth questioning; detail enhancement specifically involves automatically inserting richer detailed descriptions into the descriptive parts of the original text, which are derived from knowledge graphs related to the topic of the original text; background expansion specifically involves dynamically inserting background knowledge links or extended reading materials related to the content at appropriate positions in the text; and in-depth questioning specifically involves adding in-depth questions aimed at promoting critical thinking at the end of paragraphs.

[0017] As a preferred technical solution of this application, in step five, the AI ​​content generation model is a large language model based on the Transformer architecture and fine-tuned by the text style transfer task. It receives the original text and the adjustment instructions from step four as input, and outputs the transformed text that meets the requirements of the instructions and retains the core semantics of the original text.

[0018] As a preferred technical solution of this application, in step six, the feedback data specifically includes the user's reading fluency index and comprehension test performance when reading the converted text. The feedback data serves as a reward signal for reinforcement learning and is used to fine-tune the parameters of the AI ​​content generation model online. The feedback data is also used to update the user ability profile in the reading ability assessment model and calibrate the difficulty judgment criteria of the text difficulty analysis model.

[0019] As a preferred technical solution of this application, in step seven, the reading materials in the e-book resource library are dynamically calibrated and updated based on the reading behavior feedback data of all users through a text difficulty analysis model.

[0020] In summary, this application includes at least one of the following beneficial technical effects of a real-time simplification and enhancement method for e-book content based on reading ability assessment:

[0021] This application, by sensing users' reading behavior in real time, can dynamically assess their reading ability and accurately match it with the difficulty of the text. This allows for the automatic simplification of texts to reduce cognitive load when users encounter difficulties, and the enhancement of content to stimulate deeper thinking when users are at ease. This not only significantly improves the fluency and depth of reading, but also achieves continuous adaptation of reading materials to users' ability levels through a closed-loop optimization system and personalized recommendations. Ultimately, this achieves the fundamental goal of teaching according to aptitude and gradually improving readers' reading abilities. Attached Figure Description

[0022] Figure 1 This is a flowchart of the method for real-time simplification and enhancement of e-book content in this application. Detailed Implementation

[0023] The following is in conjunction with the appendix Figure 1 This application will be described in further detail.

[0024] See Figure 1 A method for real-time simplification and enhancement of e-book content based on reading ability assessment, characterized by the following steps:

[0025] Step 1: Collect multimodal behavioral data of users during the reading process in real time. Multimodal behavioral data includes reading speed, page dwell time, repeated reading behavior for specific words or sentences, frequency and duration of new word clicks, and accuracy and reaction time of embedded comprehension tests.

[0026] The first step is to build a comprehensive and efficient data collection system. This system needs to be seamlessly embedded into the e-book reading platform and be able to continuously and accurately capture a series of preset multimodal behavioral indicators without interfering with the user's normal reading.

[0027] The system needs to accurately record the speed changes of users reading different chapters by calculating the number of characters read per unit time or the page-turning interval. Simultaneously, the system needs to time the duration of user dwell time on each page or paragraph to determine the time required for information digestion. To identify repetitive reading behavior, the system needs to track whether users repeatedly browse specific words or sentences, which may manifest as multiple cursor stops in the same area or eye gaze retracing. For vocabulary learning behavior, the system needs to record the frequency of users clicking on vocabulary annotations and the duration of each definition viewing, reflecting their cognitive load on unfamiliar words. Furthermore, the system needs to embed short comprehension quiz questions at key points in the text and simultaneously record the user's accuracy rate and the reaction time from question presentation to answer submission. All this behavioral data needs to be timestamped in real time and linked to specific text content to form a complete behavioral data stream, providing a solid data foundation for subsequent analysis.

[0028] To achieve seamless and accurate data collection, the system needs to integrate a client-side monitoring module. This module utilizes the reader application's API hooks to listen for and record user interaction events, such as clicks, scrolling, page turning, and text selection events, and assigns them high-precision timestamps and text position anchors (such as character offsets). For new word clicks, click events of specific CSS classes are listened for and timed. For repeated reading behavior, the relationship between scroll bar position sequences and timestamps is analyzed, combined with custom algorithms (such as detecting back-and-forth scrolling or paragraphs with abnormally long dwell times within specific text areas) for identification. Embedded tests are predefined in JSON format and inserted into the text's DOM nodes, and the system records the user's choices and reaction times through event listeners. All data, after initial encapsulation, is sent to the backend service in real time via a WebSocket connection.

[0029] Step two involves using the multimodal behavioral data collected in step one to dynamically construct and update the user's real-time reading ability profile using a reading ability assessment model. The reading ability profile is a multi-dimensional vector containing vocabulary mastery, syntactic comprehension, information processing speed, and background knowledge familiarity. Specifically, step two includes: the reading ability assessment model adopts a time-series-based recurrent neural network model, taking the user's historical reading behavior data sequence as input and outputting a multi-dimensional vector; the reading ability profile is dynamically updated at configurable time intervals to ensure that it reflects the user's latest reading ability status.

[0030] The reading ability assessment model is explicitly defined as a time-series-based recurrent neural network model. Its implementation process first requires organizing the user's historical reading behavior data collected in step one into a regular data sequence in chronological order, serving as the model's input. The recurrent neural network model processes this time-dependent data, learning the ability evolution patterns corresponding to user behavior patterns. The model's output is a multi-dimensional vector that precisely quantifies the user's level in four dimensions: vocabulary mastery, syntactic comprehension, information processing speed, and background knowledge familiarity. To ensure the timeliness of the profile, the system needs to automatically trigger the model's inference process at configurable time intervals (e.g., after reading a chapter or accumulating a certain reading time), dynamically updating this multi-dimensional vector using the latest behavioral data sequence. This ensures that the generated reading ability profile always reflects the user's latest and most accurate ability status and that the updated results are persistently stored.

[0031] The reading ability assessment model can be implemented using a Long Short-Term Memory (LSTM) network. The input sequence of the model needs to be preprocessed: the heterogeneous behavioral data (such as reading speed, dwell time, test accuracy, etc.) collected in step one are aggregated according to a fixed time window (such as every 30 seconds) or by text paragraph, and normalized to the [0,1] interval to form a standardized feature vector sequence. The output layer of the model consists of four neurons, corresponding to the four ability dimensions, and the output value is limited to the (0,1) range using the Sigmoid activation function. The model needs to be pre-trained on a large, labeled user reading behavior dataset, where the labels can be provided by expert evaluation or standardized reading comprehension test scores. During online updates, the system inputs the latest behavioral data sequence into the trained model, performs forward propagation inference, and updates the user profile vector.

[0032] Step 3: Perform real-time analysis on the currently presented e-book text content. Calculate the real-time difficulty feature value of the text using a text difficulty analysis model. The real-time difficulty feature value includes lexical complexity, sentence length and structural complexity, and semantic density.

[0033] The real-time, automated analysis of the currently presented e-book text content is accomplished by a standalone text difficulty analysis model. This model needs to process the incoming text stream instantly and calculate a set of predefined real-time difficulty feature values. These feature values ​​must include lexical complexity, which may be achieved by analyzing the level, frequency, or length of words in the text; sentence length and structural complexity, which requires parsing syntactic structure to measure the complexity and semantic density of sentences, involving assessing the amount of information carried per unit length of text. The model needs to be able to quickly calculate these features for any given text segment and output a feature value vector representing its overall difficulty, while the entire process must meet low latency requirements to keep up with the user's reading pace and provide timely input for subsequent matching decisions.

[0034] The text difficulty analysis model can employ a lightweight model combining rules and statistical features to ensure real-time performance. Lexical complexity can be quantified by calculating the distribution ratio of words in the text across different difficulty levels (such as the vocabulary levels of the Modern Chinese Dictionary or the AFI graded vocabulary list). Sentence structure complexity can be evaluated by calculating indicators such as average sentence length, number of clauses, and depth of dependency syntax trees. Semantic density can be estimated by calculating indicators such as noun density and content word ratio. These feature extractors will work in parallel, and finally, a linear regression or simple neural network model (such as a multilayer perceptron MLP) will weight and fuse these features to output a comprehensive difficulty score. The weights of this fusion model are obtained by training on an existing standard corpus of manually labeled difficulty (such as a primary and secondary school Chinese textbook database).

[0035] Step four involves calculating the matching degree between the real-time reading ability profile generated in step two and the real-time difficulty feature value obtained in step three. When the matching degree is lower than the first preset threshold, a content simplification instruction is generated; when the matching degree is higher than the second preset threshold, a content enhancement instruction is generated. Step four also includes: the matching degree calculation uses a cosine similarity algorithm to calculate the similarity between the reading ability profile vector and the real-time difficulty feature value vector; the first and second preset thresholds are personalized based on the user's age group or reading ability baseline.

[0036] The system calculates the matching degree between the real-time reading ability profile (a multi-dimensional vector) generated in step two and the real-time difficulty feature value (another multi-dimensional vector) obtained in step three. The calculation method is explicitly specified as the cosine similarity algorithm. The system first needs to perform necessary preprocessing on the two vectors (such as normalization), and then calculate the cosine similarity between them to obtain a quantified matching degree score. Subsequently, the system needs to compare this score with two preset thresholds: a first preset threshold and a second preset threshold. These thresholds need to be personalized based on the user's demographic attributes (such as age group) or their initial reading ability baseline. When the calculated matching degree is lower than the first preset threshold, the system should generate a content simplification instruction; when the matching degree is higher than the second preset threshold, a content enhancement instruction should be generated. This comparison and decision-making logic needs to be executed through a stable rule engine.

[0037] Before calculating the matching degree, the two vectors need to be standardized, for example, using Z-score standardization to make their mean 0 and standard deviation 1, so as to eliminate the influence of different units. The cosine similarity is calculated using the standard formula: similarity = (A·B) / (||A|| * ||B||), where A and B are the standardized ability profile vector and difficulty feature vector, respectively. The personalized setting of the first preset threshold (e.g., 0.3) and the second preset threshold (e.g., 0.8) can be achieved by dividing users into different groups (e.g., "beginner readers" and "intermediate readers") and preset a verified threshold mapping table for each group.

[0038] Step 5: In response to the content simplification or enhancement instructions generated in Step 4, the corresponding AI content generation model is invoked to perform real-time conversion of the original text. In Step 5, the content simplification instructions include vocabulary simplification, syntactic simplification, and insertion of explanations. Specifically, vocabulary simplification involves selecting simple words from a predefined thesaurus that match the user's vocabulary proficiency and automatically replacing complex words in the original text. Syntactic simplification involves using natural language processing technology to automatically segment long and complex sentences, simplify clause structures, or adjust word order to reduce sentence complexity. Insertion of explanations involves dynamically inserting floating text explanations or illustrations next to complex concepts or words. Step 5 also includes content enhancement instructions, such as detail enhancement. The steps include: background expansion and in-depth questioning; detail enhancement specifically involves automatically inserting richer detailed descriptions into the descriptive parts of the original text, with these details derived from knowledge graphs related to the original text's theme; background expansion specifically involves dynamically inserting background knowledge links or extended reading materials related to the content at appropriate locations in the text; in-depth questioning specifically involves adding deeper questions at the end of paragraphs to promote critical thinking; in step five, the AI ​​content generation model is a large-scale language model based on the Transformer architecture and fine-tuned through a text style transfer task. It receives the original text and adjustment instructions from step four as input, and outputs the transformed text that meets the requirements of the instructions while maintaining the core semantics of the original text.

[0039] This application utilizes a designated AI content generation model, described as a large-scale language model based on the Transformer architecture and fine-tuned through a text style transfer task. Upon receiving the content simplification or enhancement instruction from step four, along with the original text, the model performs real-time text conversion according to the instruction. For content simplification instructions, the model performs three specific operations: lexical simplification (replacing complex words in the original text with simpler words from a predefined thesaurus); syntactic simplification (segmenting long sentences, simplifying clauses, or adjusting word order using natural language processing techniques); and explanation insertion (dynamically generating and inserting floating-window explanations next to complex concepts). For content enhancement instructions, the model performs three additional operations: detail enhancement (automatically inserting rich details from relevant knowledge graphs); background expansion (dynamically inserting relevant background knowledge links or materials); and in-depth questioning (adding questions that promote critical thinking at the end of paragraphs). Regardless of the instruction, the model must ensure that the converted text conforms to the requirements of the instruction while maintaining the core semantics of the original text.

[0040] The system encapsulates the conversion instructions, the original text fragments, and related constraints (such as "preserve core semantics" and "target reading level: L2") into a clear prompt word template. For example, the prompt word for the simplification instruction might be: "Simplify the following text to the sixth grade level: [Original text]. Requirements: 1. Replace complex words with simple synonyms; 2. Break down long sentences; 3. Add a one-sentence explanation for '[complex concept]'." The model (such as an LLaMA or ChatGLM with fine-tuned instructions) generates the converted text based on this prompt word. The thesaurus (such as the extended version of the Thesaurus) and knowledge graph (such as CN-DBpedia) serve as external knowledge sources and can be dynamically referenced by the prompt word through the API interface, for example, "Replace 'hesitant' with a synonym marked as 'common' in the thesaurus."

[0041] Step Six: Collect feedback data from users reading the text transformed in Step Five and use this feedback data to optimize the reading ability assessment model, text difficulty analysis model, and AI content generation model, forming an adaptive closed-loop optimization system. Specifically, in Step Six, the feedback data includes the user's reading fluency index and comprehension test performance when reading the transformed text. The feedback data serves as a reward signal for reinforcement learning, used to fine-tune the parameters of the AI ​​content generation model online. The feedback data is also used to update the user ability profile in the reading ability assessment model and calibrate the difficulty judgment criteria of the text difficulty analysis model.

[0042] The system needs to collect feedback data from users again after reading the new text transformed in step five. This data includes reading fluency metrics (such as new reading speed and dwell time) and comprehension test performance (new accuracy and reaction time). This feedback data is then used to optimize the three core models in the system: First, the feedback data serves as a reward signal in reinforcement learning, used to fine-tune the parameters of the AI ​​content generation model online, making its future text transformations more effective. Second, this new behavioral data is fed back to the reading ability assessment model to update the user's ability profile, making it more accurate. Finally, the reading feedback data from all users on various types of text is aggregated to dynamically calibrate the difficulty judgment criteria of the text difficulty analysis model, making its assessment of text difficulty more consistent with reality. Through this process, the system achieves continuous self-optimization based on user feedback.

[0043] The reward function can be defined as: Reward = w1 * fluency improvement rate + w2 * comprehension improvement rate.

[0044] Among them, the fluency improvement rate is the normalized ratio of reading speed of new and old texts, and the comprehension improvement rate is a comprehensive indicator of the improvement in test accuracy and the reduction in reaction time; w1 and w2 are adjustable weights; based on this reward signal, reinforcement learning algorithms such as Proximal Policy Optimization (PPO) can be used to fine-tune the AI ​​content generation model online; for the reading ability assessment model, new data is directly used to update the user's input data sequence and re-infer; the calibration of the text difficulty analysis model is achieved by collecting a large number of "user ability profile-reading feedback" data pairs and periodically retraining the difficulty fusion model to make its predicted difficulty more consistent with the user's actual reading experience.

[0045] Step 7: Based on real-time reading ability profiles, actively select and recommend reading materials with matching difficulty from the e-book resource library to construct personalized reading sequences. The recommendation system continuously optimizes its recommendation strategy based on user feedback on the recommended materials. In Step 7, the reading materials in the e-book resource library are dynamically calibrated and updated based on the reading behavior feedback data of all users through a text difficulty analysis model.

[0046] Based on the real-time reading ability profile generated in step two, the system needs to proactively select reading materials from a central e-book resource library whose text difficulty matches the user's current ability. The difficulty labels of the materials in this resource library are not fixed, but are dynamically calibrated and updated by the text difficulty analysis model in step three based on the reading behavior feedback data of all users mentioned in step six, thereby ensuring the accuracy of the difficulty score. The recommendation system will construct a personalized reading sequence for the user and continuously optimize its internal recommendation strategy algorithm based on the user's reading feedback on the recommended materials (such as completion rate, liking, and comprehension), thereby achieving increasingly accurate reading material recommendations and promoting the gradual development of the user's ability.

[0047] Each book in the e-book resource library undergoes model preprocessing in step three to obtain its difficulty feature vector. During recommendation, the system calculates the cosine similarity between the user's current reading ability profile vector and the difficulty vectors of all books in the resource library, and returns the Top-K books with the highest similarity. The optimization of the recommendation strategy is reflected in the fact that the system records the user's clicks, completion rate, and feedback ratings for recommended books. Using this feedback data, the system dynamically adjusts the weights of difficulty matching and interest matching in the final recommendation score through collaborative filtering or logistic regression models, thereby achieving personalized ranking.

[0048] This application, by sensing users' reading behavior in real time, can dynamically assess their reading ability and accurately match it with the difficulty of the text. This allows for the automatic simplification of texts to reduce cognitive load when users encounter difficulties, and the enhancement of content to stimulate deeper thinking when users are at ease. This not only significantly improves the fluency and depth of reading, but also achieves continuous adaptation of reading materials to users' ability levels through a closed-loop optimization system and personalized recommendations. Ultimately, this achieves the fundamental goal of teaching according to aptitude and gradually improving readers' reading abilities.

[0049] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for real-time simplification and enhancement of e-book content based on reading ability assessment, characterized in that, Includes the following steps: Step 1: Collect multimodal behavioral data of users during the reading process in real time. Multimodal behavioral data includes reading speed, page dwell time, repeated reading behavior for specific words or sentences, frequency and duration of new word clicks, accuracy and reaction time of embedded comprehension test; Step 2: Based on the multimodal behavioral data collected in Step 1, a reading ability assessment model is used to dynamically construct and update the user's real-time reading ability profile. The reading ability profile is a multi-dimensional vector that includes vocabulary mastery, syntactic comprehension, information processing speed, and background knowledge familiarity. Step 3: Perform real-time analysis on the currently presented e-book text content, and calculate the real-time difficulty feature value of the text through a text difficulty analysis model. The real-time difficulty feature value includes lexical complexity, sentence length and structural complexity, and semantic density. Step 4: Calculate the matching degree between the real-time reading ability profile generated in Step 2 and the real-time difficulty feature value obtained in Step 3. When the matching degree is lower than the first preset threshold, generate a content simplification instruction; when the matching degree is higher than the second preset threshold, generate a content enhancement instruction. Step 5: In response to the content simplification or content enhancement instructions generated in Step 4, call the corresponding AI content generation model to perform real-time conversion of the original text; Step six: Collect feedback data from users reading the text transformed in step five, and use the feedback data to optimize the reading ability assessment model, text difficulty analysis model, and AI content generation model to form an adaptive closed-loop optimization system; Step 7: Based on real-time reading ability profiles, actively select and recommend reading materials with matching difficulty from the e-book resource library to build personalized reading sequences. The recommendation system continuously optimizes its recommendation strategy based on user feedback on the recommended materials.

2. The method for real-time simplification and enhancement of e-book content based on reading ability assessment according to claim 1, characterized in that, Step two specifically includes: The reading ability assessment model uses a time-series-based recurrent neural network model, taking the user's historical reading behavior data sequence as input and outputting a multi-dimensional vector; The reading ability profile is dynamically updated at configurable time intervals to ensure it reflects the user's latest reading ability status.

3. The method for real-time simplification and enhancement of e-book content based on reading ability assessment according to claim 1, characterized in that, Step four also includes: The matching degree calculation uses the cosine similarity algorithm to calculate the similarity between the reading ability profile vector and the real-time difficulty feature value vector; The first and second preset thresholds are set individually based on the user's age group or reading ability baseline.

4. The method for real-time simplification and enhancement of e-book content based on reading ability assessment according to claim 1, characterized in that, In step five, the content simplification instructions include lexical simplification, syntactic simplification, and insertion of explanations; Specifically, vocabulary simplification involves selecting simple words from a predefined thesaurus that match the user's vocabulary proficiency and automatically replacing complex words in the original text. Syntactic simplification specifically involves using natural language processing technology to automatically segment long and complex sentences, simplify clause structures, or adjust word order to reduce sentence structural complexity. The "insert explanation" feature involves dynamically inserting floating text explanations or illustrations next to complex concepts or terms.

5. The method for real-time simplification and enhancement of e-book content based on reading ability assessment according to claim 1, characterized in that, In step five, content enhancement instructions include detail enhancement, background expansion, and in-depth questioning; The detail enhancement specifically involves automatically inserting richer detailed descriptions into the descriptive sections of the original text. These details are derived from knowledge graphs related to the theme of the original text. Background expansion specifically involves dynamically inserting relevant background knowledge links or extended reading materials at appropriate locations within the text. In-depth questioning specifically involves adding deeper questions at the end of paragraphs that aim to promote critical thinking.

6. The method for real-time simplification and enhancement of e-book content based on reading ability assessment according to claim 1, characterized in that, In step five, the AI ​​content generation model is a large language model based on the Transformer architecture and fine-tuned by a text style transfer task. It receives the original text and the adjustment instructions from step four as input, and outputs the transformed text that meets the requirements of the instructions and retains the core semantics of the original text.

7. The method for real-time simplification and enhancement of e-book content based on reading ability assessment according to claim 1, characterized in that, In step six, specifically, the feedback data includes the user's reading fluency index and comprehension test performance when reading the converted text. The feedback data serves as a reward signal for reinforcement learning and is used to fine-tune the parameters of the AI ​​content generation model online. The feedback data is also used to update the user ability profile in the reading ability assessment model and calibrate the difficulty judgment criteria of the text difficulty analysis model.

8. The method for real-time simplification and enhancement of e-book content based on reading ability assessment according to claim 1, characterized in that, In step seven, the reading materials in the e-book resource library are dynamically calibrated and updated based on reading behavior feedback data from all users through a text difficulty analysis model.

Images