Generative AI-based parent-child role playing partner training system

Abstract

The invention provides a parent-child role playing partner training system based on generative AI, and relates to the technical field of data processing, and the method comprises the steps: an action output module which synthesizes the input characteristics of a user, the achievement degree of the exercise target and the current state of the user, generates and outputs the speech content of a partner training person, outputs a corresponding field, and outputs the speech content of the partner training person; the current situation, user input and practice progress are accurately judged, and the output direction is accurately grasped; the personalized information summarization module is used for summarizing personalized information characteristics, including output preference and strategy preference, of the user according to the dialogue record after each exercise is finished; and the personalized information updating module is used for combining the current summarized personalized information with the historical personalized information and updating the current summarized personalized information and the historical personalized information into the current latest personalized information of the user so as to form long-term memory of the user. According to the invention, the user can be guided to learn more efficiently.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control technology, and in particular to a parent-child role-playing training system based on generative AI. Background Art

[0002] With the rapid development of science and technology, artificial intelligence (AI) technology has been widely used in various fields. Especially in the field of education, the application of AI technology provides learners with a more personalized and efficient learning experience. Parent-child education, as an important part of education, also benefits from the advancement of AI technology. However, most of the existing parent-child education auxiliary systems focus on content provision and interactive entertainment, and lack in-depth training functions for specific practice goals.

[0003] Traditional parent-child role-playing training usually relies on the experience and knowledge of parents or teachers, so it is difficult to ensure the professionalism and sustainability of the training. In addition, since each child has different learning characteristics and progress, some traditional training methods cannot meet the needs of personalized learning. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a parent-child role-playing training system based on generative AI, which can guide users to learn more efficiently.

[0005] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0006] First, the parent-child role-playing training system based on generative AI includes:

[0007] The user input observation module is used to observe the historical dialogue records each time the user inputs, and analyze the characteristics of the user input and the degree of achievement of the exercise goal according to the user's current input;

[0008] The user status thinking module determines the user's current status based on pre-set practice goals, relevant practice points, knowledge points and other materials, including practice points that have been achieved and practice points that have not been achieved;

[0009] The action output module combines the characteristics of the user input, the degree of achievement of the practice goal, and the user's current status to generate and output the speech content of the sparring partner, and outputs the corresponding fields, including accurate judgment of the current situation, user input, and practice progress, as well as accurate grasp of the output direction;

[0010] Personalized information summary module: After each exercise, the personalized information characteristics of the user are summarized based on the conversation records, including output preferences and strategy preferences;

[0011] The personalized information update module merges the currently summarized personalized information with the historical personalized information and updates it to the latest user personalized information to form the user's long-term memory.

[0012] Furthermore, each time the user inputs, the historical dialogue records are observed, and according to the user's current input, the characteristics of the user's input and the degree of achievement of the practice goal are analyzed, including:

[0013] Whenever a user inputs something, the user's input content is stored in a historical conversation record, which includes a timestamp, the user input text, and context information at the time of input;

[0014] Preprocess the user's current input and extract features from the preprocessed text. The features include word frequency, n-gram, TF-IDF value and semantic vector;

[0015] Compare the features of the user's current input with the data in the historical conversation records to analyze changes in the user's language style, topic selection, and emotional tendencies;

[0016] Use semantic analysis to extract semantic features of user input, including keywords, entities, sentiment and sentence meaning, etc.; compare the extracted semantic features with the evaluation rules to calculate the scores of user input on various evaluation indicators;

[0017] According to the scores of each evaluation indicator, the comprehensive score of the user input is calculated, and the comprehensive score reflects the overall achievement of the exercise goal.

[0018] Furthermore, the calculation formula for the comprehensive score input by the user is:

[0019]

[0020] Among them, C S Represents the comprehensive score input by the user; N represents the total number of evaluation indicators; i represents the index of the evaluation indicator, which is used to traverse all evaluation indicators; represents the Gaussian weight function; μ represents the mean of the Gaussian function; σ represents the standard deviation of the Gaussian function; α i , β i , γ i and δ i represents weight; U i represents the number of unique words in the i-th evaluation metric; T i represents the total number of words in the i-th evaluation index; X i represents the number of complex sentences in the i-th evaluation index; Y i represents the total number of sentences in the i-th evaluation index; v j-1 and v jrepresents the adjacent sentences in the text; N represents the number of evaluation indicators; L i Indicates the length of the input text; L max Indicates the preset maximum length; n represents the number of sentences in the text.

[0021] Furthermore, based on the pre-set practice goals, relevant practice points, knowledge points and other materials, the user's current status is judged, including the practice points that have been achieved and the practice points that have not been achieved, including:

[0022] Determine pre-set practice goals, and collect user performance data during practice through practice activities, testing, and user feedback, including user answer records, accuracy, reaction time, and error types;

[0023] Compare the collected user data with the practice points and knowledge points to identify the key points;

[0024] Based on the key points, determine the user's current status.

[0025] Furthermore, the characteristics of the user input, the degree of achievement of the practice goal, and the user's current state are comprehensively considered to generate and output the speech content of the sparring partner, and the corresponding fields are output at the same time, including accurate judgment of the current situation, user input, and practice progress, as well as accurate grasp of the output direction, including:

[0026] Analyze user input, including language expression, selected answers, and response time; identify keywords, tone, and emotions in user input to understand user intent and attitude;

[0027] Compare the user input with the exercise goal to evaluate the user's achievement in this exercise;

[0028] According to the user's current status and practice achievement, construct the speech content that the sparring partner needs to provide, including encouragement, guidance, error correction and suggestions;

[0029] The constructed training speech content is output to the user, and the corresponding fields are output, as well as annotations of the judgment of the current situation, the evaluation of the user input, the overview of the progress of the practice, and the grasp of the output direction.

[0030] The second aspect is the parent-child role-playing training method based on generative AI, including:

[0031] Every time the user inputs, the historical conversation records are observed, and the characteristics of the user's input and the degree of achievement of the exercise goal are analyzed according to the user's current input;

[0032] According to the pre-set practice goals, relevant practice points, knowledge points and other materials, the user's current status is judged, including the practice points that have been achieved and the practice points that have not been achieved;

[0033] Based on the characteristics of the user input, the degree of achievement of the practice goal, and the user's current status, the speech content of the sparring partner is generated and output, and the corresponding fields are output at the same time, including accurate judgment of the current situation, user input, and practice progress, as well as accurate grasp of the output direction;

[0034] After each exercise, summarize the user's personalized information characteristics based on the conversation records, including output preferences and strategy preferences;

[0035] The currently summarized personalized information is merged with the historical personalized information and updated to the current latest user personalized information to form the user's long-term memory.

[0036] According to a third aspect, a computing device includes:

[0037] one or more processors;

[0038] The storage device is used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method described.

[0039] In a fourth aspect, a computer-readable storage medium stores a program, and when the program is executed by a processor, the method described is implemented.

[0040] The above solution of the present invention includes at least the following beneficial effects:

[0041] The user input observation module analyzes user input in real time, and the system can quickly identify the learner's learning characteristics and problems, so as to adjust the training strategy in time, provide targeted guidance, and effectively improve learning efficiency. The user status thinking module continuously tracks the learner's learning status to ensure that the learner does not miss any important knowledge points or skill points during the practice process, and comprehensively improves the learning effect.

[0042] The action output module generates personalized speech content based on the learner's real-time status and input characteristics, ensuring that each learner can obtain training support that fits their needs and enhance their learning motivation. The personalized information summary module and the update module work together to accumulate learners' personalized data for the system, enabling the system to meet learners' personalized needs more and more accurately over time.

[0043] The system is based on parent-child role-playing and uses generative AI to intelligently train parents, which promotes effective interaction between parents and children and deepens the emotional connection between parents and children. The training speech content generated by the system takes into account the characteristics of parent-child communication and helps to create a positive and harmonious learning atmosphere.

[0044] When using the system, learners need to actively input, reflect and adjust their learning strategies, which helps to cultivate their autonomous learning and problem-solving abilities. The timely feedback and guidance provided by the system can help learners establish correct learning methods and habits. As an intelligent tutoring assistant, the system can greatly reduce the burden on parents during tutoring, allowing parents to participate in their children's learning more easily. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a schematic diagram of a parent-child role-playing training system based on generative AI provided by an embodiment of the present invention.

[0046] Figure 2 It is a flow chart of a parent-child role-playing training method based on generative AI provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0048] like Figure 1 As shown, an embodiment of the present invention proposes a parent-child role-playing sparring system based on generative AI, including:

[0049] The user input observation module is used to observe the historical dialogue records each time the user inputs, and analyze the characteristics of the user input and the degree of achievement of the exercise goal according to the user's current input;

[0050] The user status thinking module determines the user's current status based on pre-set practice goals, relevant practice points, knowledge points and other materials, including practice points that have been achieved and practice points that have not been achieved;

[0051] The action output module combines the characteristics of the user input, the degree of achievement of the practice goal, and the user's current status to generate and output the speech content of the sparring partner, and outputs the corresponding fields, including accurate judgment of the current situation, user input, and practice progress, as well as accurate grasp of the output direction;

[0052] Personalized information summary module: After each exercise, the personalized information characteristics of the user are summarized based on the conversation records, including output preferences and strategy preferences;

[0053] The personalized information update module merges the currently summarized personalized information with the historical personalized information and updates it to the latest user personalized information to form the user's long-term memory.

[0054] In an embodiment of the present invention, the system can quickly identify the learning characteristics and problems of learners by analyzing user input in real time through the user input observation module, so as to adjust the training strategy in time, provide targeted guidance, and effectively improve learning efficiency. The user status thinking module continuously tracks the learning status of learners to ensure that learners do not miss any important knowledge points or skill points during the practice process, and comprehensively improve the learning effect. The action output module generates personalized speech content according to the real-time status and input characteristics of the learners, ensuring that each learner can obtain training support that fits their needs and enhance learning motivation. The personalized information summary module and the update module work together to accumulate learners' personalized data for the system, so that the system can meet learners' personalized needs more and more accurately over time. The system is based on parent-child role-playing, and through the intelligent training of generative AI, it promotes effective interaction between parents and children and deepens the emotional connection between parents and children. The training speech content generated by the system takes into account the characteristics of parent-child communication and helps to create a positive and harmonious learning atmosphere. In the process of using the system, learners need to actively input, reflect and adjust their learning strategies, which helps to cultivate their autonomous learning ability and problem-solving ability. The timely feedback and guidance provided by the system can help learners establish correct learning methods and habits. As an intelligent training assistant, the system can greatly reduce the burden of parents during the training process, allowing parents to participate in their children's learning more easily.

[0055] In the embodiment of the present invention, each time the user inputs, the historical dialogue record is observed, and according to the current input of the user, the characteristics of the user input and the degree of achievement of the practice goal are analyzed, including:

[0056] Whenever a user inputs, the user's input content is stored in the historical conversation record. The historical conversation record includes the timestamp, the user input text, and the context information at the time of input. Specifically, it includes: creating a database table or data structure for storing historical conversation records. The structure should contain fields such as: user ID, timestamp, input text, context information (such as conversation turn, previous input, etc.); setting a listener in the system, and the listener is triggered whenever the user inputs; after the listener captures the user input, it stores the input content, current timestamp and context information together in the data structure of the historical conversation record.

[0057] Preprocess the user's current input, and extract features from the preprocessed text. The features include word frequency, n-gram, TF-IDF value, and semantic vector. Specifically, it includes: removing irrelevant characters and punctuation marks, using Regular Expression to identify and remove special characters, non-standard characters, and punctuation marks in the input text; loading a predefined stop word list (such as commonly used words like "的", "是", "在", etc., which contribute little to semantics), traversing the input text, and removing the stop words that appear in the text. Use Chinese word segmentation tools (such as jieba, THULAC, etc.) to segment the preprocessed text. The word segmentation tool can split continuous text into a sequence of meaningful words, facilitating subsequent feature extraction and analysis.

[0058] Word frequency statistics:

[0059] Initialize an empty dictionary to store word frequency information. Traverse the sequence of segmented words. For each word, check if it already exists in the dictionary. If it exists, increment the count of that word by 1; if it does not exist, add the word to the dictionary and initialize the count to 1. The resulting dictionary is the word frequency statistics result.

[0060] n-gram extraction:

[0061] Define the value of n (such as 2 or 3). Traverse the sequence of segmented words, and extract n-gram features in a sliding window manner. Store the extracted n-gram features in a set or list to remove duplicates.

[0062] TF-IDF value calculation:

[0063] TF (term frequency) calculation: directly use the previously calculated word frequency as the TF value; IDF (inverse document frequency) calculation: a corpus is required to calculate the IDF value of each word. The specific steps are as follows:

[0064] Count the total number of documents in the corpus. For each word, count how many documents it appears in;

[0065] Use the formula IDF(t) = log(D / dt) to calculate the IDF value, where D is the total number of documents, and dt is the number of documents containing the word t; multiply the TF value of each word by its IDF value to obtain the TF-IDF value of that word.

[0066] Semantic vector generation:

[0067] Load a pre-trained word vector model (such as Word2Vec, BERT, etc.). For each segmented word, find its corresponding vector representation from the word vector model. There are various ways to combine these word vectors to obtain the semantic vector of the entire text. For example:

[0068] Average all word vectors.

[0069] Use the TF-IDF value as the weight to perform weighted averaging on the word vectors.

[0070] If you use a model such as BERT, you can directly input the entire sentence and get the embedding vector of the sentence.

[0071] The final vector is the semantic vector representation of the text.

[0072] Compare the features of the user's current input with the data in the historical conversation records to analyze changes in the user's language style, topic selection, and emotional tendency, including:

[0073] According to the user ID, all historical input data of the user is retrieved from the historical conversation record database, ensuring that the retrieved data contains the user input text, timestamp, and necessary context information. According to the analysis requirements, historical data within a specific time period can be selectively filtered and the retrieved historical data can be sorted in chronological order for subsequent time series analysis.

[0074] Preprocess the user's current input (such as removing irrelevant characters, word segmentation, etc.) and extract features (word frequency, n-gram, TF-IDF value, semantic vector), using the same methods and tools as before to ensure consistency. Compare the word frequency of the current input with the average word frequency of the historical input or the word frequency of a specific historical period to observe changes in vocabulary usage habits. Compare the use of n-grams in the current input with that in historical data to analyze the continuity and changes in user phrase usage. Compare the similarities and differences between the current input and important words (high TF-IDF values) in historical data to understand the shift in user focus. Use methods such as cosine similarity to calculate the similarity between the current input semantic vector and the historical input semantic vector, and analyze the consistency and deviation of user topics through changes in similarity.

[0075] Based on the comparison results of word frequency and n-gram, the changes in the use of user vocabulary and phrases are analyzed to determine the evolution of language style (such as formality, colloquialism, etc.). By comparing TF-IDF values ​​and semantic vectors, the relevance and difference between the user's current input and historical input in terms of topics are identified, and the trend of user topic selection is analyzed by observing the continuation, conversion or emergence of new topics.

[0076] Select a sentiment analysis tool based on the needs and the context of the analysis, such as a rule-based approach (e.g., sentiment dictionary), a machine learning model (e.g., a trained sentiment classifier), or a deep learning model (e.g., a Transformer-based sentiment analysis model). Integrate the selected sentiment analysis tool into the existing system to ensure that it can smoothly process user input and return sentiment analysis results. Perform necessary preprocessing on the user's current input, such as removing irrelevant characters and word segmentation, to meet the requirements of the sentiment analysis tool. Pass the preprocessed user input to the sentiment analysis tool. The tool performs sentiment analysis on the user input based on the built-in rules or models, and returns a sentiment tendency label (e.g., positive, neutral, negative) and the corresponding confidence or score.

[0077] Retrieve the user's historical input data from the historical conversation records, and perform the same preprocessing steps as the current input on each historical input. Pass the preprocessed historical input data in batches to the sentiment analysis tool. The tool performs sentiment analysis on each historical input and returns the corresponding sentiment tendency label and score. Aggregate the sentiment analysis results of the current input with the sentiment analysis results of the historical input to form a time series data set. Each data contains a timestamp, input text, sentiment tendency label, and score. Analyze the time series data set to observe the changing trend of sentiment tendency:

[0078] The frequency of positive, neutral, and negative emotions in different time periods was counted, and statistical indicators such as the mean and standard deviation of the emotion scores were calculated to quantify the amplitude of emotion fluctuations.

[0079] Analyze the fluctuation of users’ emotional state according to the changes in emotional tendencies:

[0080] Identify time periods and events with greater emotional fluctuations, and explore the causes of emotional fluctuations, such as external stimuli, personal needs, etc.

[0081] Based on the trend of emotional tendency changes in historical data, try to predict the user's future emotional state:

[0082] Use time series forecasting models (such as ARIMA, LSTM, etc.) to predict sentiment scores.

[0083] Semantic analysis is used to extract semantic features of user input, including keywords, entities, sentiment tendencies, and sentence meanings. The extracted semantic features are compared with the evaluation rules to calculate the comprehensive score of the user input, which reflects the overall achievement of the exercise objectives.

[0084] In an embodiment of the present invention, by storing the content and context information of each user input, the system can establish a complete historical conversation record. This helps to fully trace the user's learning process and provide rich materials for subsequent data analysis and optimization. By preprocessing and extracting features of the user's current input, the system can deeply understand the user's language style, topic preference and emotional tendency. This understanding helps the system to grasp user needs more accurately and provide personalized training services. By comparing the features of the user's current input with historical data, the system can promptly discover the changes and progress of the user in the learning process. This helps the system dynamically adjust the training strategy to ensure that it always keeps pace with the user's learning needs. By using semantic analysis technology to extract the semantic features of the user's input and comparing them with the evaluation rules, the system can accurately calculate the user's score on each evaluation indicator. This provides users with objective and quantitative learning feedback, which helps them determine their learning progress and the direction that needs to be improved. By calculating the comprehensive score of the user's input, the system can comprehensively measure the user's overall achievement of the practice goal. This helps the system to give more targeted learning suggestions and guidance, helping users achieve their practice goals more efficiently.

[0085] In this embodiment of the present invention, the calculation formula for the comprehensive score input by the user is:

[0086]

[0087] Among them, C S Represents the comprehensive score input by the user; N represents the total number of evaluation indicators; i represents the index of the evaluation indicator, which is used to traverse all evaluation indicators; represents the Gaussian weight function; μ represents the mean of the Gaussian function; σ represents the standard deviation of the Gaussian function; α i , β i , γ i and δ i represents weight; U i represents the number of unique words in the i-th evaluation metric; T i represents the total number of words in the i-th evaluation index; X i represents the number of complex sentences in the i-th evaluation index; Y i represents the total number of sentences in the i-th evaluation index; v j-1 and v j represents the adjacent sentences in the text; N represents the number of evaluation indicators; L i Indicates the length of the input text; L max Indicates the preset maximum length; n represents the number of sentences in the text.

[0088] In the embodiment of the present invention, the formula comprehensively considers multiple evaluation indicators, including the number of unique words, the number of complex sentence patterns, the coherence between sentences, and the length of the text, so as to comprehensively evaluate the quality of the user input. This comprehensiveness ensures that all aspects of the user input can be quantitatively evaluated, improving the accuracy and objectivity of the evaluation. By introducing the Gaussian weight function and different weight coefficients α i , β i , γ i and δ i , the formula can flexibly adjust the weight of each indicator according to different evaluation needs. This makes the formula adaptable to different application scenarios and evaluation focuses. The introduction of the number of unique words and complex sentence patterns in the formula allows the evaluation to not only stay at the surface level, but also go deep into the semantic and syntactic structure of the text. This refined evaluation method helps to more accurately reflect the quality and level of user input. Measuring the coherence of the text by calculating the cosine similarity between adjacent sentences is an innovative and practical method. It can not only evaluate the quality of a single sentence, but also consider the logic and reading fluency of the text as a whole. By introducing the ratio of the text length to the preset maximum length as one of the evaluation indicators, the formula can control the length of the user input to a certain extent, avoiding the influence of too long or too short text on the overall evaluation. This helps to guide users to input more refined and effective content.

[0089] In the embodiment of the present invention, the current status of the user is judged based on the pre-set practice goals, relevant practice points, knowledge points and other materials, including the practice points that have been achieved and the practice points that have not been achieved, including:

[0090] Determine pre-set practice goals, and collect user performance data during the practice process through practice activities, tests and user feedback, including the user's answer record, accuracy, reaction time and error type. Specifically, according to the teaching content or training needs, determine the specific goals of the practice, such as "actively listen to children in conversations, understand and empathize with their feelings"; design corresponding practice activities and test questions according to the practice goals to ensure that they can effectively evaluate whether the user has achieved the practice goals; when the user is performing practice activities and tests, the user's answer situation is recorded through the system background, including the answer result of each question (correct or incorrect), reaction time (the time from the display of the question to the user submitting the answer), and error type (such as grammatical errors, calculation errors, comprehension errors, etc.); at the same time, user feedback data can also be collected, such as the user's perception of the difficulty of the practice, the degree of understanding of the knowledge points, etc., which can be achieved through questionnaires, user interviews or feedback forms after the practice.

[0091] Compare the collected user data with the practice points and knowledge points to identify key points, including: organize the collected user data, classify them according to the practice points and knowledge points, and form a clear data structure; compare the user's answer records with the pre-set practice points and knowledge points one by one, and mark the user's performance on each point and knowledge point; pay special attention to the user's error type and reaction time, as this information helps to identify the specific links where the user has problems; and identify the points and knowledge points that the user has mastered well, as well as the points and knowledge points that the user has difficulty or misunderstanding, by analyzing user data.

[0092] Based on the key points, judge the user's current status, including: according to the user's performance on the key points, evaluate their current learning status, such as "mastered", "partially mastered", "not mastered", etc., and set some quantitative standards to help with the evaluation, such as "mastered" when the accuracy rate reaches 80%, "partially mastered" when it reaches 60%-80%, and "not mastered" when it is lower than 60%. The evaluation results are presented in the form of a report, which should list in detail the practice points that the user has achieved and the practice points that have not been achieved. The report can also include an analysis of the user's error types, statistics on reaction time, and specific suggestions for each unmastered point. Provide personalized feedback and guidance suggestions based on the user's current status. For the knowledge points that have been mastered, affirmation and encouragement can be given; for the knowledge points that have been partially mastered or not mastered, targeted learning resources and practice suggestions are provided to encourage users to conduct subsequent learning and practice based on feedback and guidance to continuously improve their ability level.

[0093] In an embodiment of the present invention, by analyzing user data, the system can provide personalized learning suggestions for each user to help them achieve their practice goals more effectively. Real-time data collection and analysis enable the system to provide instant feedback, allowing learners to understand their progress and areas for improvement. This instant feedback can motivate learners and help them adjust their learning strategies in a timely manner. By accurately identifying the key points of practice that have been achieved and not achieved, learners can avoid wasting time on unnecessary content and instead focus on overcoming their weaknesses, thereby significantly improving learning efficiency. Teachers or coaches can allocate teaching resources and attention more reasonably based on the data provided by the system, and provide more targeted guidance to learners. When learners see their progress and goals achieved, they will gain a sense of accomplishment, which will enhance their motivation and confidence in learning. Traditional evaluation methods may rely on a single test score, but by continuously collecting and analyzing user data, a more scientific and comprehensive learning effect evaluation can be obtained, which helps to more accurately measure learners' actual abilities and progress.

[0094] In the embodiment of the present invention, the characteristics of the user input, the degree of achievement of the exercise goal and the current state of the user are comprehensively considered to generate and output the speech content of the sparring partner, and at the same time output the corresponding fields, including accurate judgment of the current situation, user input and exercise progress, and accurate grasp of the output direction, including:

[0095] Analyze the user's input, including language expression, selected answers and reaction time; identify keywords, tone and emotions in user input to understand the user's intentions and attitudes, including: obtaining answers or feedback given by users through text, voice, multiple-choice questions, etc.; performing pre-processing operations such as word segmentation and stop word removal on the collected text; using algorithms such as TF-IDF and TextRank to extract keywords in user input; using natural language processing technology to identify the tone in user input, such as affirmation, negation, question, etc.; using sentiment analysis models to determine the emotions in user input, such as happiness, frustration, anxiety, etc.; recording the user's reaction time from seeing the question to giving the answer; if the user answers in the form of multiple-choice questions, analyze the answer options selected by the user; combining the above information to understand the user's intentions and attitudes, such as learning attitude, self-confidence, etc.

[0096] Compare the user input with the exercise goal and evaluate the user's achievement in this exercise, including: determining the specific goal of this exercise, comparing the user's answer with the standard answer to determine its correctness, and calculating the user's achievement in this exercise based on factors such as the user's accuracy rate and reaction time. If the user's answer is wrong, identify the type of error, such as poor grasp of knowledge points, misunderstanding, etc., analyze which questions or knowledge points the user performed poorly on, identify their learning difficulties, compare user historical data, analyze the user's progress and regression in the current exercise, and combine the above information to comprehensively evaluate the user's performance in this exercise.

[0097] According to the user's current status and practice achievement, the speech content that the coach needs to provide is constructed. The speech content includes encouragement, guidance, correction of errors and suggestions. Specifically, according to the user's performance, the tone of the coaching speech is determined, such as encouragement, guidance, correction, etc. If the user performs well, encouraging words are written to enhance their self-confidence. For the problems encountered by the user in learning, specific guidance and suggestions are provided. For the user's wrong answers, the correct answers are pointed out in a gentle manner and personalized learning suggestions are formulated according to the user's personal characteristics and needs. The above-mentioned types of speech content are integrated to form a coherent coaching speech. According to the user's preferences and situational needs, the language style of the coaching speech is adjusted. The constructed coaching speech is reviewed and modified to ensure its accuracy and appropriateness.

[0098] Output the constructed training discourse content to the user, and output the corresponding fields, as well as mark the judgment of the current situation, the evaluation of the user input, the overview of the practice progress and the grasp of the output direction, specifically including: determine the output method of the training discourse, such as text, voice, etc., output the constructed training discourse to the user in the selected way, and output the field information related to the training discourse, such as user ID, practice ID, timestamp, etc., mark the current learning situation in the output, such as practice stage, difficulty level, etc., include the evaluation of the user input in the output, such as accuracy, reaction speed, etc., summarize the user's progress in this practice, ensure that the output training discourse is consistent with the practice goal, guide the user in the right direction, save each output record for subsequent analysis and improvement, and collect user feedback on the training discourse after output for continuous optimization.

[0099] In an embodiment of the present invention, by accurately analyzing the user's input content, including language expression, selected answers and reaction time, the trainer can more accurately understand the user's intentions and attitudes. Based on this information, the trainer can output more targeted encouraging words, thereby effectively improving the user's participation and motivation for continuous practice. By identifying the keywords, tone and emotions in the user's input, the trainer can provide personalized guidance and suggestions. When the user makes a mistake, the trainer can correct it immediately and give specific improvement directions, which helps the user to master the knowledge points and improve skills faster. By comparing the user input with the practice goal, the trainer can objectively evaluate the user's achievement in this practice. This instant feedback mechanism allows users to clearly understand their progress and areas for improvement, and then adjust their learning strategies. According to the user's current state and practice achievement, the trainer can construct appropriate speech content to guide the user along the optimized learning path. This not only improves learning efficiency, but also avoids users wasting time on invalid or inefficient learning directions. The speech content of the trainer is intended to provide a positive learning experience, helping users build confidence and reduce frustration in the learning process through encouragement, guidance and suggestions. This humanized interactive method makes users feel cared for and supported, so that they can enjoy the learning process more. By marking the judgment of the current situation, the evaluation of the user's input, the overview of the practice progress and the grasp of the output direction, the sparring partner can ensure that each output is targeted, which not only meets the user's current needs, but also helps to promote the achievement of practice goals.

[0100] After each practice session, collect and analyze the conversation records, including the user's language expression, selected answers, reaction time, conversation turns, emotional changes, and other information. Analyze the user's response to the practice speech, such as whether there is a more positive feedback to a certain type of encouragement or advice. Identify the user's acceptance and preference for different output methods (such as text, voice, charts, etc.). Record the user's preference for the level of information detail, such as whether they prefer concise answers or detailed explanations.

[0101] Analyze the learning strategies adopted by users during practice, such as step-by-step reasoning, direct guessing, and elimination. Examine users’ correction behavior for incorrect answers to determine whether they tend to self-correct or rely on external guidance. Pay attention to users’ behavioral patterns when faced with difficult questions, such as whether they tend to ask for help, give up, or keep trying. Extract key features from conversation records, such as users’ language style, common vocabulary, and expressions. Count users’ behavioral patterns during practice, such as answering speed, frequency of asking for help, etc.

[0102] The results of the above analysis are integrated to form a personalized information characteristics report for the user. The report includes the user's output preferences (such as preferred communication methods, information presentation forms) and strategy preferences (such as learning style, problem-solving strategies).

[0103] Retrieve the user's historical personalized information from the user profile or database. Compare the current information with the historical information:

[0104] Compare and analyze the currently summarized personalized information with the historical records to identify the consistency and difference and determine whether the user's preference has changed; if the current information is consistent with the historical information, strengthen the weight of the preference; if the current information shows a new preference or change, merge it with the historical information to form a more comprehensive user portrait; for contradictory information, weigh and choose according to the frequency of occurrence and recent behavior patterns. The weight calculation formula is:

[0105]

[0106] Among them, FW i represents the final weight, which means the weight of the user's preference after comprehensive consideration of historical data and current data; N i Indicates the number of times the preference appears in the historical records, which reflects the user's historical tendency towards preference; N total represents the total number of times all preferences appear in historical records. This is a normalization factor used to calculate the relative frequency of each preference in historical data. α represents the consistency influence coefficient, which is a small positive number used to adjust the influence of the consistency factor on the final weight. When the current information is consistent with the historical information, this coefficient determines the extent of the weight increase. C countIndicates the number of times the current information is consistent with the historical information; β indicates the change impact coefficient, which is used to adjust the influence of the change factor on the final weight and reflects the system's sensitivity to changes in user preferences; M i Indicates the number of times the preference occurs in the current cycle; M total Indicates the total number of times all preferences appear in the current cycle.

[0107] The merged personalized information is organized into the user's latest personalized profile, which contains key information such as the user's long-term preferences, learning style, and problem-solving strategies. The updated personalized information is stored back in the user profile or database, and data is backed up regularly to prevent information loss. The latest personalized information is fed back to the intelligent practice system so that the system can adjust the practice content, difficulty, and practice style according to the user's personalized characteristics.

[0108] like Figure 2 As shown, an embodiment of the present invention also provides a parent-child role-playing training method based on generative AI, including:

[0109] Every time the user inputs, the historical conversation records are observed, and the characteristics of the user's input and the degree of achievement of the exercise goal are analyzed according to the user's current input;

[0110] According to the pre-set practice goals, relevant practice points, knowledge points and other materials, the user's current status is judged, including the practice points that have been achieved and the practice points that have not been achieved;

[0111] Based on the characteristics of the user input, the degree of achievement of the practice goal, and the user's current status, the speech content of the sparring partner is generated and output, and the corresponding fields are output at the same time, including accurate judgment of the current situation, user input, and practice progress, as well as accurate grasp of the output direction;

[0112] After each exercise, summarize the user's personalized information characteristics based on the conversation records, including output preferences and strategy preferences;

[0113] The currently summarized personalized information is merged with the historical personalized information and updated to the current latest user personalized information to form the user's long-term memory.

[0114] For example, when it comes to specific applications, the specific implementation process is as follows:

[0115] There are three major steps for users to conduct AI smart training, namely: training scene creation (before training), practice (during practice), and (after training) evaluation and personalized information extraction (after practice). The following is an explanation of the relevant parts in the above figure:

[0116] 1. Training scenario creation:

[0117] Brief description: Generate targeted practice scenarios (including: role information of the sparring partner corresponding to the ability points, major conflict events, etc.) based on the practitioner's personalized information, previous practice evaluations, and the practice topics (ability points) selected by the user.

[0118] Input: ability point name, user's personalized information, user's previous practice evaluation;

[0119] Output: Practice scenario: including role information, conflict events, event background, etc.

[0120] Internal operation details: The following steps are performed to obtain the generated personalized sparring scenario:

[0121] Obtain the relevant definition, description and professional resources (such as "thinking model") of the capability through the provided capability name;

[0122] Query the user's personalized information and past related skill practice evaluations from the database through the user ID;

[0123] Generate scene information through "scene generation agent";

[0124] Main output information:

[0125] time_location: the time and place where the event occurred, such as "Thursday night, home";

[0126] background: event background, background information of the conflict event, such as: "The child was misunderstood by the teacher at school, the teacher tore up the child's homework book, and the child was very sad";

[0127] trigger_event: trigger event: triggers the plot in which the user role-playing communicates with the AI ​​role-playing, for example: "The child comes home, says nothing, and leaves his room";

[0128] 2. Practice:

[0129] Brief description: After having a practice scenario, the user starts role-playing with the AI ​​to practice the selected ability points. The AI ​​will communicate with the user based on its role and the scenario information generated in the previous step.

[0130] Input: Practice all the conversations so far (for this training scenario), practice ability points, and user personalized information.

[0131] Output: The AI ​​character’s response to the user, including: the current conversation content, emotional state (changes), psychological activities, and suggestions for the user’s handling method (suggestions are based on the “thinking model”).

[0132] Internal operation details: Through the following steps, AI will output language, emotions, and psychological activities that are suitable for the current conversation context and can serve as a sparring partner:

[0133] The training agent observes the current conversation and analyzes the current user status and characteristics (without specifying the specific analysis dimensions, only requiring a summary of the current conversation);

[0134] The training agent thinks about the next strategy: it obtains relevant information about the practice ability point, including the definition, introduction and ability model ("thinking model"), and requires the large language model to determine: i) which step of the ability model it is currently at, and ii) what is the most appropriate response as a training partner (response strategy).

[0135] The training agent executes the strategy generated in the previous step, that is, based on the above reaction strategy and combined with the user's personalized information (extracted from the database), it gives the dialogue content, emotional state and inner activities that conform to the character's characteristics; and outputs some additional information to help the user better organize the next round of dialogue. The main information output is shown in the following section.

[0136] The main output information:

[0137] "role_feature": role features. The AI ​​will infer the character's features based on the scenario during the first conversation and will update them continuously in the future.

[0138] "role_limitation": role limitation, mainly the insufficiency of AI's role-playing ability, such as: average language expression ability, poor expression of emotions, in order to better simulate real parent-child conversations

[0139] "hit_the_model": The value is "yes" or "no", indicating whether the user's latest conversation content conforms to a certain step of the "thinking model".

[0140] "mood_change": mood change, the value is "better" or "worse", the impact of the user's current input on the mood of the AI ​​character.

[0141] "thoughts": the psychological activities of the character played by AI at this time,

[0142] "words": words that the AI ​​character wants to say,

[0143] "emotion": the emotional state of the character played by the AI ​​at this time,

[0144] "current_step": the step of the "thinking model" the user is currently taking

[0145] "suggested_step": recommends the user to take the next step in the "thinking model",

[0146] "progress": The current progress, which is based only on the completion rate of the steps specified by the thinking model, with a value of "0%" to "100%".

[0147] 3. Evaluation and personalized information extraction:

[0148] Based on the entire conversation during the practice, we extract personalized information about the user and evaluate the effectiveness of the practice, provide feedback for the practice, and enable the practitioner to reflect and improve.

[0149] Input: the entire practice conversation; previously extracted user personalized information; practice ability points.

[0150] Output: User practice process and result evaluation (completion scores and summary suggestions based on the thinking model); merged user personalized information.

[0151] The following steps are involved:

[0152] Evaluation Agent Observation: Summarize the user's behavioral characteristics from the conversation (the direction and dimension of the summary are not specified, and LLM is required to summarize on its own).

[0153] Evaluate Agent thinking: Compare the thinking model (extracted according to the name of the capability point) and analyze the completion quality for each step (or dimension), including the good and bad parts of the completion.

[0154] Evaluation Agent Action: Based on the above results, score the user's final practice process and provide summary, suggestions and demonstrations.

[0155] Personalized information Agent extracts information: Summarizes the user's personalized characteristics from the conversation, including: response strategy, response mode and other personalized information.

[0156] Personalized information agent merges information: extracts the user's most recent personalized information from the database and merges it with the currently extracted personalized information. The large language model is required to merge the two pieces of information. If there is a conflict in the information, the new information shall prevail.

[0157] The evaluation results and new personalized information are stored in the corresponding database.

[0158] Main output information:

[0159] evaluate:

[0160] For each step (or each dimension) in the thinking model, output:

[0161] "keypint": step / dimension name

[0162] "bonus_behavior": bonus items, that is, points that users can perform to get bonus points in this step or dimension;

[0163] "deduction_behavior": deduction item, the point that the user should be deducted for in this step or dimension;

[0164] "score": the dimension score of this step, 0-10 points;

[0165] “evidence”: the evidence (reason) for the above score;

[0166] For the overall evaluation of the practice process, the following will be output:

[0167] “summary”: a summary of the practice process;

[0168] “suggestions”: Suggestions for unfinished items in the thinking model;

[0169] "demonstration": demonstration, some typical demonstrations of unfinished steps;

[0170] Personalized information:

[0171] Extraction stage: The three fields of response strategy, response preference and other personalized information need to be output

[0172] Merge phase: merge the old personalized information with the new personalized information, and also output the contents of the above three fields.

[0173] It should be noted that the system is a system corresponding to the above method, and all implementation methods in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0174] The embodiment of the present invention further provides a computing device, comprising: a processor, a memory storing a computer program, wherein when the computer program is executed by the processor, the method described above is executed. All implementations in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0175] The embodiment of the present invention also provides a computer-readable storage medium storing instructions, which, when executed on a computer, enable the computer to execute the method described above. All implementations in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0176] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A parent-child role-playing sparring system based on generative AI, characterized by: include: The user input observation module is used to observe the historical dialogue records each time the user inputs, and analyze the characteristics of the user input and the degree of achievement of the exercise goal according to the user's current input; The user status thinking module determines the user's current status based on pre-set practice goals, relevant practice points, knowledge points and other materials, including practice points that have been achieved and practice points that have not been achieved; The action output module combines the characteristics of the user input, the degree of achievement of the practice goal, and the user's current status to generate and output the speech content of the sparring partner, and outputs the corresponding fields, including accurate judgment of the current situation, user input, and practice progress, as well as accurate grasp of the output direction; Personalized information summary module: After each exercise, the personalized information characteristics of the user are summarized based on the conversation records, including output preferences and strategy preferences; The personalized information update module merges the currently summarized personalized information with the historical personalized information and updates it to the latest user personalized information to form the user's long-term memory.

2. The parent-child role-playing training system based on generative AI according to claim 1 is characterized in that: Each time the user inputs, the historical dialogue records are observed, and based on the user's current input, the characteristics of the user's input and the degree to which it achieves the exercise goal are analyzed, including: Whenever a user inputs something, the user's input content is stored in a historical conversation record, which includes a timestamp, the user input text, and context information at the time of input; Preprocess the user's current input and extract features from the preprocessed text. The features include word frequency, n-gram, TF-IDF value and semantic vector; Compare the features of the user's current input with the data in the historical conversation records to analyze changes in the user's language style, topic selection, and emotional tendencies; Use semantic analysis to extract semantic features of user input, including keywords, entities, sentiment and sentence meaning, etc.; compare the extracted semantic features with the evaluation rules to calculate the scores of user input on various evaluation indicators; According to the scores of each evaluation indicator, the comprehensive score of the user input is calculated, and the comprehensive score reflects the overall achievement of the exercise goal.

3. The parent-child role-playing training system based on generative AI according to claim 2 is characterized in that: The formula for calculating the comprehensive score entered by the user is: Among them, C S Represents the comprehensive score input by the user; N represents the total number of evaluation indicators; i represents the index of the evaluation indicator, which is used to traverse all evaluation indicators; represents the Gaussian weight function; μ represents the mean of the Gaussian function; σ represents the standard deviation of the Gaussian function; α i , β i , γ i and δ i represents weight; U i represents the number of unique words in the i-th evaluation metric; T i represents the total number of words in the i-th evaluation index; X i represents the number of complex sentences in the i-th evaluation index; Y i represents the total number of sentences in the i-th evaluation index; v j-1 and v j represents the adjacent sentences in the text; N represents the number of evaluation indicators; L i Indicates the length of the input text; L max Indicates the preset maximum length; n represents the number of sentences in the text.

4. The parent-child role-playing training system based on generative AI according to claim 3 is characterized in that: According to the pre-set practice goals, relevant practice points, knowledge points and other materials, the user's current status is judged, including the practice points that have been achieved and the practice points that have not been achieved, including: Determine pre-set practice goals, and collect user performance data during practice through practice activities, testing, and user feedback, including user answer records, accuracy, reaction time, and error types; Compare the collected user data with the practice points and knowledge points to identify the key points; Based on the key points, determine the user's current status.

5. The parent-child role-playing training system based on generative AI according to claim 4 is characterized in that: Based on the characteristics of the user input, the degree of achievement of the practice goal, and the user's current status, the speech content of the sparring partner is generated and output, and the corresponding fields are output, including accurate judgment of the current situation, user input, and practice progress, as well as accurate grasp of the output direction, including: Analyze user input, including language expression, selected answers, and response time; identify keywords, tone, and emotions in user input to understand user intent and attitude; Compare the user input with the exercise goal to evaluate the user's achievement in this exercise; According to the user's current status and practice achievement, construct the speech content that the sparring partner needs to provide, including encouragement, guidance, error correction and suggestions; The constructed training speech content is output to the user, and the corresponding fields are output, as well as annotations of the judgment of the current situation, the evaluation of the user input, the overview of the progress of the practice, and the grasp of the output direction.

6. A parent-child role-playing sparring method based on generative AI, characterized in that: Applicable to the system according to any one of claims 1 to 5, comprising: Every time the user inputs, the historical conversation records are observed, and the characteristics of the user's input and the degree of achievement of the exercise goal are analyzed according to the user's current input; According to the pre-set practice goals, relevant practice points, knowledge points and other materials, the user's current status is judged, including the practice points that have been achieved and the practice points that have not been achieved; Based on the characteristics of the user input, the degree of achievement of the practice goal, and the user's current status, the speech content of the sparring partner is generated and output, and the corresponding fields are output at the same time, including accurate judgment of the current situation, user input, and practice progress, as well as accurate grasp of the output direction; After each exercise, summarize the user's personalized information characteristics based on the conversation records, including output preferences and strategy preferences; The currently summarized personalized information is merged with the historical personalized information and updated to the current latest user personalized information to form the user's long-term memory.

7. A computing device, characterized in that include: one or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method according to claim 6.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program, which implements the method according to claim 6 when executed by a processor.

Images