An AI-based tablet learning and training system and method

Abstract

The present invention discloses a tablet computer learning and training system and method based on artificial intelligence, which relates to the field of tablet computer applications and includes: inputting a preset learning material data set into the local database of the tablet computer; extracting the text in the learning material data set and performing standardization processing to obtain a learning material text data set, and performing named entity recognition through a first machine learning model; performing relationship extraction on the named entity data set to obtain named entity relationship classification, performing knowledge fusion based on the semantic relationship of text features and adding attribute labels and relationship labels to establish a learning material knowledge graph; identifying the facial micro-expression image data of the user through a second machine learning model to obtain a user learning state recognition result, calculating the distraction frequency and performing threshold discrimination, and the tablet computer executes a message prompt based on the threshold discrimination result; establishing a learning and training model, and setting an initial course, a derivative course, a derivative training question group and a personalized training question group for the user.

Description

Technical Field

[0001] The present invention relates to the technical field of tablet computer applications, and particularly to a tablet computer learning and training system and method based on artificial intelligence. Background Art

[0002] Thanks to the rapid development of electronic technology and industry, the market of tablet computers has been growing steadily and the penetration rate has been continuously increasing. In the education industry, the application of tablet computers as new teaching aids has become a normal trend. Tablet computers can support multimedia learning, presenting teaching knowledge from multiple angles of text, images, and videos, and can improve interactive operations, enhancing learners' understanding and memory. Tablet computers are portable and convenient to carry around for learning during fragmented time. With the rise of artificial intelligence, tablet computers have provided a new direction in the field of learning and training.

[0003] Currently, the Chinese invention patent with the application number CN202311773999.9 discloses a method and system for realizing an education platform based on a tablet computer. Specifically, the invention includes: obtaining the capability information and index information of the tablet computer; collecting the keyword information of the user learning history of the learning applications on the tablet computer; the education platform determines the division granularity of the keywords based on the capability information, index information, and keyword information; classifies and associates all the learning resources on the education platform according to the division granularity; obtains the similarity between the target user or target learning resource and the existing learning resources on the education platform; and recommends learning resources based on the similarity. By classifying learning resources according to keywords, the present invention can more accurately understand and organize the content of learning resources; determines the division granularity according to the performance and storage capacity of the tablet computer and makes flexible adjustments according to specific situations. However, when integrating learning materials, this invention uses a fixed algorithm and lacks adaptability to different types of materials. When classifying knowledge points, it only relies on keyword similarity and lacks the division and recommendation of learning materials from the perspectives of attributes and relationships. The degree of personalization in arranging user courses and training questions is insufficient, and there is a lack of interaction with users, failing to effectively judge the learning status of users and issue prompt messages according to the judgment results. Summary of the Invention

[0004] The technical problem solved by the present invention is that traditional methods use a fixed algorithm when integrating learning material data using tablet computers, lacking adaptability to different types of materials. It fails to establish a material graph from the perspectives of the attributes and relationships of learning material named entities. At the same time, the degree of personalization in arranging user courses and training questions is insufficient, and there is a lack of interaction with users, failing to effectively judge the learning status of users and issue graded prompt messages according to the judgment results.

[0005] To solve the above technical problems, the present invention provides the following technical solutions:

[0006] A tablet computer learning and training method based on artificial intelligence, comprising:

[0007] Step S1, inputting preset video course data, textbook graphic and text data, and training question data into the local database of the tablet computer to obtain a learning material data set;

[0008] Step S2, extracting the text in the learning material data set and performing standardization processing to obtain a learning material text data set, and performing named entity recognition on the learning material text data set through a first machine learning model to obtain a named entity data set;

[0009] Step S3, performing relation extraction on the named entity data set to obtain a named entity relation classification result, generating a text feature semantic relation based on the named entity relation classification result, performing knowledge fusion based on the text feature semantic relation, adding attribute labels and relation labels to the learning material data set, and establishing a learning material knowledge graph through a Neo4j graph database;

[0010] Step S4, collecting user's facial micro-expression image data through the front camera of the tablet computer, performing recognition on the user's facial micro-expression image data through a second machine learning model to obtain a user learning state recognition result, calculating the distraction frequency based on the user learning state recognition result and performing threshold discrimination, and the tablet computer executing a first message prompt and a second message prompt based on the threshold discrimination result;

[0011] Step S5, establishing a learning and training model, and setting an initial course, a derivative course, a derivative training question group, and a personalized training question group based on the subject selected by the user, the learning material knowledge graph, the historical error rate of answering questions, and the distraction frequency.

[0012] As a preferred solution of the tablet computer learning and training method based on artificial intelligence according to the present invention, wherein: extracting the video course name index, textbook text, and training questions in the learning material data set, performing standardization processing in text format to obtain a learning material text data set, and performing named entity recognition on the learning material text data set through a first machine learning model to obtain a named entity data set;

[0013] The named entity data set includes a concept entity set, a question entity set, an answer entity set, a solution entity set, a person entity set, an event entity set, and a time entity set.

[0014] As a preferred solution of a tablet learning and training method based on artificial intelligence according to the present invention, wherein: the first machine learning model is a pre-trained improved BERT named entity model, and the logic of the first machine learning model for identifying the learning material text data set is as follows: the learning material text data set is subjected to word embedding and segment embedding through an embedding layer to obtain an extended data set, the text statements of the extended data set are vectorized and marked to obtain a text vector set, the text vector set is weighted through an attention mechanism to obtain an updated vector set, a text vector sequence is obtained through non-linear transformation and vector splicing of the updated vector set, bidirectional feature extraction is performed on the text vector sequence through a memory network in a hidden layer, feature extraction is performed on the text vector sequence according to the forward word order and the reverse word order respectively, a text feature vector sequence is obtained through vector splicing, probability annotation is performed on the text feature vector sequence through a conditional random field algorithm in an output layer, and classification and decoding are completed through the joint probability of the text feature vector sequence to obtain a named entity data set.

[0015] As a preferred solution of a tablet learning and training method based on artificial intelligence according to the present invention, wherein: relation extraction is performed based on the named entity data set to obtain a text feature semantic relation, knowledge fusion is performed based on the text feature semantic relation, and a learning material knowledge graph is established through a Neo4j graph database;

[0016] The processing logic of the relation extraction is as follows: graph nodes are generated based on the named entity data set, each named entity in the named entity data set is mapped to a node, a text node graph is generated by connecting the nodes based on the text vector sequence, weighted calculation based on a graph attention mechanism is performed through a gated unit, the text node state is updated to obtain an updated text node graph, and a named entity relation classification result is obtained through global graph pooling and classification processing of the updated text node graph;

[0017] The named entity relation classification result includes: causal relation, inclusion relation, derivation relation, opposition relation, equivalence relation, temporal relation, and spatial relation.

[0018] As a preferred solution of a tablet learning and training method based on artificial intelligence according to the present invention, wherein: the processing logic of knowledge fusion is as follows: entity alignment processing is performed on the named entity data set through a Word2Vec semantic similarity algorithm to obtain an aligned named entity data set, relation fusion is performed based on the named entity relation classification result, similar relations are merged to obtain a named entity fusion relation, the aligned named entity data set and the named entity fusion relation are subjected to format unification processing according to the Neo4j data import format, and attribute labels and relation labels are added to each named entity and corresponding relation based on the aligned named entity data set and the named entity fusion relation to generate a learning material knowledge graph;

[0019] The attribute tags include subject index, primary importance, secondary importance, primary difficulty, secondary difficulty, and tertiary difficulty;

[0020] The relationship tags include causal relationship, inclusion relationship, derivation relationship, opposition relationship, equivalence relationship, time sequence relationship, and spatial relationship.

[0021] As a preferred solution of the artificial intelligence-based tablet learning and training method described in the present invention, wherein: the learning state of the user is identified and a message prompt is given to the user. The facial micro-expression image data of the user is collected through the front camera of the tablet, and the second machine learning model analyzes the facial micro-expression image data of the user to obtain the user learning state recognition result. The user learning state recognition result includes a focused state and a distracted state;

[0022] The distracted frequency per unit time is calculated based on the user learning state recognition result, and threshold discrimination is performed based on the distracted frequency. The threshold discrimination logic is: when the distracted frequency is greater than or equal to the first concentration threshold, the tablet executes the first message prompt; when the distracted frequency is greater than or equal to the second concentration threshold, the tablet executes the second message prompt;

[0023] The first message prompt is: the tablet microphone plays the first prompt sound effect and emits vibrations, and at the same time, the screen display brightness changes uniformly back and forth between the original brightness and the maximum brightness. The duration of the first message prompt is 2 seconds;

[0024] The second message prompt is: the tablet microphone plays the second prompt sound effect, stops the learning and training, and pops up a rest notice.

[0025] As a preferred solution of the artificial intelligence-based tablet learning and training method described in the present invention, wherein: the second machine learning model is a pre-trained improved Yolov5 model. The logic for the second machine learning model to analyze the facial micro-expression image of the user is: at the input end, Mosaic is adopted for image data enhancement, and the facial micro-expression image data of four consecutive video frames is fused to obtain a fused image;

[0026] A Focus module is introduced into the backbone network to perform adjacent domain downsampling processing on the fused image, and interval sampling and segmentation convolution processing are performed on the pixels of the fused image to obtain preliminary image features;

[0027] Introduce the BiFPN structure into the neck network, add weights to the input values of each channel in the preliminary image features respectively, and obtain shallow fusion features and deep fusion features through weighted calculation. Establish cross-scale bidirectional connections between convolutional layers, transfer the shallow fusion features to the deep layer and add position information to the deep fusion features, transfer the deep fusion features to the shallow layer and add semantic information to the shallow fusion features, and further perform feature fusion processing through convolutional layers to obtain weighted fusion image features;

[0028] Use a decoupled detection head to replace the detection head in the head network, and perform bounding box prediction and class prediction on the weighted fusion image features respectively through the classifier and regressor in parallel in the decoupled detection head, and combine the non-maximum suppression algorithm to obtain the learning state recognition result;

[0029] When training the second machine learning model, the DioU-LOSS loss function is used to replace the original CioU-LOSS loss function, and the loss is measured by calculating the norm distance and overlapping area between the target box and the predicted box. The training set and validation set of the second machine learning model are obtained by dividing the preset facial micro-expression image set according to a ratio of 7:3. The preset facial micro-expression image set is a learning state image with pre-labeled tags. The facial micro-expression image set includes focused images, daze images, dozing images, yawning images, and eye shift images. The labeled image tags include focused state and distracted state.

[0030] As a preferred solution of the artificial intelligence-based tablet learning and training method described in the present invention, wherein: establish a learning and training model, set an initial course based on the subject data selected by the user, set derivative courses and derivative training question groups based on relationship tags, and set personalized training question groups based on the difficulty tags, historical error rates of answering questions, and distraction frequencies of the derivative training question groups.

[0031] As a preferred solution of the artificial intelligence-based tablet learning and training method described in the present invention, wherein: the logic for setting the initial course is: use the subject data selected by the user as the subject index in the attribute tags, search through the learning material knowledge graph to obtain the user target named entity set, extract the concept entities in the user target named entity set to obtain the user target concept entity set, and randomly extract from the user target concept entity set according to a ratio of 4:6 of the primary importance and the secondary importance based on the attribute tags of the user target concept entity set to obtain the initial course;

[0032] The logic for setting up derivative courses and derivative training question sets is as follows: Search for the derivative named entity set through the relationship tags of the user's target concept entity set, extract the question entity set from the derivative named entity set to obtain the derivative question entity set, and randomly select from the derivative question entity set according to the ratio of 3:4:3 for primary difficulty, secondary difficulty, and tertiary difficulty based on the attribute tags of the derivative question entity set to obtain the derivative training question set;

[0033] The logic for setting up personalized training question sets is as follows: Perform threshold screening on the historical error rates of answering questions for the question entities in the derivative training question set, remove the question entities with error rates less than or equal to the first error rate threshold, perform relationship tag search on the question entities with error rates greater than the second error rate threshold to obtain the error-prone named entity set, extract the question entity set from the error-prone named entity set to obtain the error-prone question entity set, and randomly select from the error-prone question entity set according to the ratio of 2:1 for secondary difficulty and tertiary difficulty based on the attribute tags of the error-prone question entity set to obtain the personalized training question set.

[0034] An artificial intelligence-based tablet learning and training system includes: a data storage module, a data processing module, a learning and training module, an interaction module, and a power module;

[0035] The data storage module is used to store the learning material dataset, the named entity dataset, the learning material knowledge graph, and the user historical data. The data storage module is connected to the learning and training module and the data processing module, receives the learning material knowledge graph data sent by the data processing module, receives the user historical data sent by the learning and training module, receives the request instructions from the learning and training module and the data processing module, and sends the corresponding stored data to the learning and training module and the data processing module;

[0036] The data processing module is used to generate the learning material knowledge graph corresponding to the learning material dataset through the first machine learning model, identify the user's facial micro-expression image through the second machine learning model to obtain the user's learning status result and calculate the message prompt data. The data processing module sends the learning material knowledge graph data and the user's learning status result data to the data storage module, and sends the message prompt data to the interaction module;

[0037] The learning and training module is used to set up the initial course, derivative courses, derivative training question sets, and personalized training question sets, provide courses and training questions for users. The learning and training module is connected to the data storage module, sends request instructions to the data storage module and receives the stored data corresponding to the instructions from the data storage module. The learning and training module is connected to the interaction module and receives the user's answer result data from the interaction module;

[0038] The interaction module is used for human-computer interaction with users, receiving the user learning status results sent by the data processing module and sending out message prompts, receiving the courses and training questions of the learning and training module and displaying them to the users, recording the user's answer results and sending them to the learning and training module, receiving user instructions and sending them to the data storage module and the data processing module. The user instructions include searching for named entities, selecting subjects, selecting courses, answering questions, and exiting the system;

[0039] The power supply module is used to provide power for the data storage module, the data processing module, the learning and training module, and the interaction module.

[0040] Advantages of the present invention: The first machine learning model efficiently extracts the named entity dataset from the learning material dataset, which is beneficial to integrating the text data of learning materials in different disciplines and categories. Establishing the learning material relationship graph is beneficial to systematically constructing a knowledge system for users, quickly, efficiently, and accurately locating the retrieval target during user retrieval. Through the named entity relationship classification and the attribute tags of the named entity dataset, the deep relationships of each knowledge point in the learning materials are displayed in the form of a structure diagram, which helps to provide support for scientifically arranging courses and questions. The second machine learning model quickly analyzes the user's facial micro-expression image data, which is beneficial to obtaining in real time whether the user is in a focused state or a distracted state. The interaction ability is enhanced through the sound and light message prompts, which is beneficial to improving the user's learning and training efficiency. The improved second machine learning model is beneficial to fully extracting the tiny features of the pupils, eyes, and mouth in the micro-expression image data. In the execution of the prediction task step during training, not only the overlapping area between the target box and the prediction box is considered, but also the Euclidean distance between the center points of the target box and the prediction box is considered, improving the detection accuracy of the user's facial micro-expression changes. Setting the user's courses and training questions according to the attribute tags, difficulty tags, historical error rate of answering questions, and distraction frequency is beneficial to targeted supplementation of the user's learning weak links and improving the level of personalized recommendation. Description of the Drawings

[0041] Figure 1 It is a schematic diagram of the basic process of a tablet computer learning and training method based on artificial intelligence provided by an embodiment of the present invention;

[0042] Figure 2 It is a schematic diagram of the basic framework of a tablet computer learning and training system based on artificial intelligence provided by an embodiment of the present invention. Detailed Embodiments

[0043] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following detailed description of the specific embodiments of the present invention is made in conjunction with the accompanying drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments.

[0044] Example 1, referring to Figure 1 , which is an embodiment of the present invention, provides an artificial intelligence-based tablet learning and training method, including:

[0045] Step S1, input preset video course data, textbook graphic data, and training question data into the local database of the tablet to obtain a learning material dataset;

[0046] Step S2, extract the text in the learning material dataset and perform standardization processing to obtain a learning material text dataset, and perform named entity recognition on the learning material text dataset through a first machine learning model to obtain a named entity dataset;

[0047] Step S3, perform relationship extraction on the named entity dataset to obtain a named entity relationship classification result, generate a text feature semantic relationship based on the named entity relationship classification result, perform knowledge fusion based on the text feature semantic relationship, add attribute labels and relationship labels to the learning material dataset, and establish a learning material knowledge graph through the Neo4j graph database;

[0048] Step S4, collect the facial micro-expression image data of the user through the front camera of the tablet, identify the facial micro-expression image data of the user through a second machine learning model to obtain a user learning state recognition result, calculate the distraction frequency based on the user learning state recognition result and perform threshold discrimination, and the tablet executes a first message prompt and a second message prompt based on the threshold discrimination result;

[0049] Step S5, establish a learning and training model, and set the initial course, derivative course, derivative training question group, and personalized training question group based on the subject selected by the user, the learning material knowledge graph, the historical error rate of answering questions, and the distraction frequency.

[0050] In this embodiment, the video course name index, textbook text, and training questions in the learning material dataset are extracted, and the learning material text dataset is obtained through text format standardization processing. Named entity recognition is performed on the learning material text dataset through a first machine learning model to obtain a named entity dataset;

[0051] The named entity dataset includes a concept entity set, a question entity set, an answer entity set, a solution entity set, a person entity set, an event entity set, and a time entity set.

[0052] In this embodiment, the first machine learning model is a pre-trained improved BERT named entity model. The logic of the first machine learning model for identifying the learning material text dataset is as follows: the learning material text dataset is subjected to word embedding and segment embedding through the embedding layer to obtain an augmented dataset. The text statements of the augmented dataset are vectorized and marked to obtain a text vector set. The text vector set is weighted through the attention mechanism to obtain an updated vector set. A text vector sequence is obtained through non-linear transformation and vector splicing of the updated vector set. The memory network in the hidden layer is used to perform bidirectional feature extraction on the text vector sequence. Feature extraction is performed on the text vector sequence in the forward word order and the reverse word order respectively. A text feature vector sequence is obtained through vector splicing. The conditional random field algorithm in the output layer is used to perform probability annotation on the text feature vector sequence. Classification and decoding are completed through the joint probability of the text feature vector sequence to obtain a named entity dataset.

[0053] In this embodiment, relationship extraction is performed based on the named entity dataset to obtain text feature semantic relationships, and knowledge fusion is performed based on the text feature semantic relationships. A learning material knowledge graph is established through the Neo4j graph database;

[0054] The processing logic of the relationship extraction is as follows: graph nodes are generated based on the named entity dataset, and each named entity in the named entity dataset is mapped to a node. A text node graph is generated by connecting each node based on the text vector sequence. Weighted calculation based on the graph attention mechanism is performed through the gated unit to update the text node state to obtain an updated text node graph. Global graph pooling and classification processing are performed on the updated text node graph to obtain a named entity relationship classification result;

[0055] The named entity relationship classification results include: causal relationship, inclusion relationship, derivation relationship, opposition relationship, equivalence relationship, temporal relationship, and spatial relationship.

[0056] In this embodiment, the processing logic of knowledge fusion is as follows: entity alignment processing is performed on the named entity dataset through the Word2Vec semantic similarity algorithm to obtain an aligned named entity dataset. Relationship fusion is performed based on the named entity relationship classification results, and similar relationships are merged to obtain a named entity fusion relationship. The aligned named entity dataset and the named entity fusion relationship are uniformly processed according to the Neo4j data import format. Attribute labels and relationship labels are added to each named entity and corresponding relationship based on the aligned named entity dataset and the named entity fusion relationship to generate a learning material knowledge graph;

[0057] The attribute labels include subject index, primary importance, secondary importance, primary difficulty, secondary difficulty, and tertiary difficulty;

[0058] The relationship tags include causal relationship, inclusion relationship, derivation relationship, opposition relationship, equivalence relationship, time sequence relationship, and spatial relationship.

[0059] The named entity dataset includes a concept entity set, a problem entity set, an answer entity set, a solution entity set, a person entity set, an event entity set, and a time entity set;

[0060] The classification results of named entity relationships include: causal relationship, inclusion relationship, derivation relationship, opposition relationship, equivalence relationship, time sequence relationship, and spatial relationship.

[0061] Among them, the named entity dataset is efficiently extracted from the learning material dataset through the first machine learning model, which is beneficial to integrating learning material data text data of different disciplines and categories. Building a learning material relationship graph is conducive to systematically constructing a knowledge system for users, quickly, efficiently, and accurately locating retrieval targets during user retrieval. Through named entity relationship classification and the attribute tags of the named entity dataset, the deep relationships of each knowledge point in the learning materials are displayed in the form of a structure diagram, which helps to provide support for scientifically arranging courses and questions.

[0062] In this embodiment, the user's learning state is identified and a message prompt is given to the user. The facial micro-expression image data of the user is collected through the front camera of the tablet computer, and the second machine learning model analyzes the facial micro-expression image data of the user to obtain the user learning state recognition result. The user learning state recognition result includes a focused state and a distracted state;

[0063] Based on the user learning state recognition result, the distraction frequency per unit time is calculated, and threshold discrimination is performed based on the distraction frequency. The threshold discrimination logic is: when the distraction frequency is greater than or equal to the first concentration threshold, the tablet computer executes the first message prompt; when the distraction frequency is greater than or equal to the second concentration threshold, the tablet computer executes the second message prompt;

[0064] The first message prompt is: the tablet computer microphone plays the first prompt sound effect and emits vibrations, and at the same time, the screen display brightness changes uniformly back and forth between the original brightness and the maximum brightness. The duration of the first message prompt is 2 seconds;

[0065] The second message prompt is: the tablet computer microphone plays the second prompt sound effect, stops the learning and training, and pops up a rest notice.

[0066] In this embodiment, the second machine learning model is a pre-trained improved Yolov5 model. The logic for the second machine learning model to analyze the user's facial micro-expression images is: at the input end, Mosaic is adopted for image data enhancement, and the facial micro-expression image data of four consecutive video frames is fused to obtain a fused image;

[0067] Introduce the Focus module into the backbone network, perform subsampling processing on the neighboring domain of the fused image, and perform interval sampling and segmentation convolution processing on the pixels of the fused image to obtain preliminary image features;

[0068] Introduce the BiFPN structure into the neck network, add weights to the input values of each channel in the preliminary image features respectively, and obtain shallow fusion features and deep fusion features through weighted calculation. Establish cross-scale bidirectional connections between convolutional layers, transfer the shallow fusion features to the deep layer and add position information to the deep fusion features, transfer the deep fusion features to the shallow layer and add semantic information to the shallow fusion features, and further perform feature fusion processing through convolutional layers to obtain weighted fusion image features;

[0069] Use a decoupled detection head to replace the detection head in the head network, and perform bounding box prediction and class prediction on the weighted fusion image features respectively through the classifier and regressor connected in parallel in the decoupled detection head, and combine the non-maximum suppression algorithm to obtain the learning state recognition result;

[0070] When training, the second machine learning model uses the DioU-LOSS loss function instead of the original CioU-LOSS loss function, and measures the loss by calculating the norm distance and overlapping area between the target box and the predicted box. The training set and validation set of the second machine learning model are obtained by dividing the preset facial micro-expression image set according to a ratio of 7:3. The preset facial micro-expression image set is a learning state image with pre-labeled tags. The facial micro-expression image set includes focused images, daze images, dozing images, yawning images, and gaze deviation images. The labeled image tags include focused state and distracted state.

[0071] Among them, analyzing the user's facial micro-expression image data quickly through the second machine learning model is beneficial to obtaining the user's focused state or distracted state in real time. The enhancement of the interaction ability through the sound and light message prompt is beneficial to improving the user's learning and training efficiency. The improved second machine learning model is beneficial to fully extracting the tiny features of the pupils, eyes, and mouth in the micro-expression image data. In the prediction task execution step during training, not only the overlapping area between the target box and the predicted box is considered, but also the Euclidean distance between the center points of the target box and the predicted box is considered, improving the detection accuracy of the user's facial micro-expression changes.

[0072] In this embodiment, a learning and training model is established. Based on the subject data selected by the user, the initial courses are set. Based on the relationship tags, the derivative courses and derivative training question groups are set. Based on the difficulty tags, the historical error rate of answering questions, and the distraction frequency of the derivative training question groups, the personalized training question groups are set.

[0073] In this embodiment, the logic for setting up the initial courses is as follows: Using the subject data selected by the user as the subject index in the attribute tags, search through the learning material knowledge graph to obtain the user's target named entity set. Extract the conceptual entities from the user's target named entity set to obtain the user's target conceptual entity set. Based on the attribute tags of the user's target conceptual entity set, randomly select from the user's target conceptual entity set according to the ratio of 4:6 for primary importance and secondary importance to obtain the initial courses;

[0074] The logic for setting up derivative courses and derivative training question groups is as follows: Search through the relationship tags of the user's target conceptual entity set to obtain the derivative named entity set. Extract the question entity set from the derivative named entity set to obtain the derivative question entity set. Based on the attribute tags of the derivative question entity set, randomly select from the derivative question entity set according to the ratio of 3:4:3 for primary difficulty, secondary difficulty, and tertiary difficulty to obtain the derivative training question groups;

[0075] The logic for setting up personalized training question groups is as follows: Perform threshold screening on the historical error rates of answering questions for the question entities in the derivative training question groups, remove the question entities with error rates less than or equal to the first error rate threshold, perform relationship tag searches on the question entities with error rates greater than the second error rate threshold to obtain the error-prone named entity set. Extract the question entity set from the error-prone named entity set to obtain the error-prone question entity set. Based on the attribute tags of the error-prone question entity set, randomly select from the error-prone question entity set according to the ratio of 2:1 for secondary difficulty and tertiary difficulty to obtain the personalized training question groups.

[0076] Among them, setting up user courses and training questions according to attribute tags, difficulty tags, historical error rates of answering questions, and distraction frequencies is conducive to specifically making up for the weak links in the user's learning and improving the level of personalized recommendation.

[0077] Example 2, referring to Figure 2 , is another embodiment of the present invention, providing an artificial intelligence-based tablet learning and training system, including: a data storage module, a data processing module, a learning and training module, an interaction module, and a power module;

[0078] The data storage module is used to store learning material data sets, named entity data sets, learning material knowledge graphs, and user historical data. The data storage module is connected to the learning and training module and the data processing module, receives the learning material knowledge graph data sent by the data processing module, receives the user historical data sent by the learning and training module, receives request instructions from the learning and training module and the data processing module, and sends the corresponding stored data to the learning and training module and the data processing module;

[0079] The data processing module is used to generate a learning material knowledge graph corresponding to the learning material dataset through the first machine learning model, identify the user's facial micro-expression image through the second machine learning model to obtain the user's learning status result, and calculate the message prompt data. The data processing module sends the learning material knowledge graph data and the user's learning status result data to the data storage module, and sends the message prompt data to the interaction module;

[0080] The learning and training module is used to set up initial courses, derivative courses, derivative training question groups, and personalized training question groups, and provide courses and training questions for users. The learning and training module is connected to the data storage module, sends a request instruction to the data storage module, and receives the stored data corresponding to the instruction from the data storage module. The learning and training module is connected to the interaction module and receives the user's answer result data from the interaction module;

[0081] The interaction module is used to conduct human-computer interaction with the user, receive the user's learning status result sent by the data processing module and issue a message prompt, receive the courses and training questions of the learning and training module and display them to the user, record the user's answer result and send it to the learning and training module, receive the user's instruction and send it to the data storage module and the data processing module. The user's instruction includes searching for named entities, selecting subjects, selecting courses, answering questions, and exiting the system;

[0082] The power supply module is used to provide power for the data storage module, the data processing module, the learning and training module, and the interaction module.

[0083] Those skilled in the art should understand that the embodiments of the present invention may provide a method, a system or a computer program product. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. Among them, the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured article including an instruction device, and the instruction device implements the process Figure 1 in one process or multiple processes and / or Figure 1 the functions specified in one block or multiple blocks.

[0084] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A learning and training method for a tablet computer based on artificial intelligence, characterized in that, Including: Step S1: Input the preset video course data, textbook graphic and text data, and training question data into the local database of the tablet computer to obtain a learning material dataset. Step S2: Extract the text in the learning material dataset and perform standardization processing to obtain a learning material text dataset. Use the first machine learning model to perform named entity recognition on the learning material text dataset to obtain a named entity dataset. Step S3: Perform relationship extraction on the named entity dataset to obtain a named entity relationship classification result. Generate a text feature semantic relationship based on the named entity relationship classification result. Perform knowledge fusion based on the text feature semantic relationship, add attribute labels and relationship labels to the learning material dataset, and establish a learning material knowledge graph through the Neo4j graph database. Step S4: Collect the user's facial micro-expression image data through the front camera of the tablet computer. Use the second machine learning model to recognize the user's facial micro-expression image data to obtain a user learning state recognition result. Calculate the distraction frequency based on the user learning state recognition result and perform threshold discrimination. The tablet computer executes the first message prompt and the second message prompt based on the threshold discrimination result. Step S5: Establish a learning and training model, and set the initial courses, derivative courses, derivative training question groups, and personalized training question groups based on the subject selected by the user, the learning material knowledge graph, the historical error rate of answering questions, and the distraction frequency. Establish a learning and training model, set the initial courses based on the subject data selected by the user, set the derivative courses and derivative training question groups based on the relationship labels, and set the personalized training question groups based on the difficulty labels of the derivative training question groups, the historical error rate of answering questions, and the distraction frequency. The logic for setting the initial courses is as follows: Use the subject data selected by the user as the subject index in the attribute labels, search through the learning material knowledge graph to obtain the user's target named entity set, extract the concept entities from the user's target named entity set to obtain the user's target concept entity set, and randomly extract from the user's target concept entity set according to the ratio of 4:6 for the primary importance and secondary importance to obtain the initial courses. The logic for setting the derivative courses and derivative training question groups is as follows: Search through the relationship labels of the user's target concept entity set to obtain a derivative named entity set, extract the problem entity set from the derivative named entity set to obtain a derivative problem entity set, and randomly extract from the derivative problem entity set according to the ratio of 3:4:3 for the primary difficulty, secondary difficulty, and tertiary difficulty to obtain the derivative training question groups. The logic for setting up the personalized training question set is as follows: perform threshold screening on the answering historical error rates of question entities in the derived training question set, remove question entities with error rates less than or equal to the first error rate threshold, conduct relationship label searches on question entities with error rates greater than the second error rate threshold to obtain an error-prone named entity set, extract the question entity set from the error-prone named entity set to obtain an error-prone question entity set, and randomly select from the error-prone question entity set according to the ratio of secondary difficulty to tertiary difficulty of 2:1 based on the attribute tags of the error-prone question entity set to obtain the personalized training question set.

2. The method for learning and training a tablet computer based on artificial intelligence according to claim 1, wherein: Extract the video course name index, teaching material text, and training questions from the learning material dataset, obtain the learning material text dataset through text format standardization processing, and perform named entity recognition on the learning material text dataset through the first machine learning model to obtain the named entity dataset; The named entity dataset includes a concept entity set, a question entity set, an answer entity set, a solution entity set, a person entity set, an event entity set, and a time entity set.

3. The method for learning and training a tablet computer based on artificial intelligence according to claim 2, characterized in that: The first machine learning model is a pre-trained improved BERT named entity model. The logic for the first machine learning model to identify the learning material text dataset is as follows: perform word embedding and segment embedding on the learning material text dataset through the embedding layer to obtain an extended dataset, perform vectorization processing on the text statements of the extended dataset and label the statements to obtain a text vector set, perform weighted processing on the text vector set through the attention mechanism to obtain an updated vector set, perform non-linear transformation and vector splicing on the updated vector set to obtain a text vector sequence, perform bidirectional feature extraction on the text vector sequence through the memory network in the hidden layer, perform feature extraction on the text vector sequence in the forward word order and reverse word order respectively, obtain a text feature vector sequence through vector splicing, perform probability annotation on the text feature vector sequence through the conditional random field algorithm in the output layer, and complete classification and decoding through the joint probability of the text feature vector sequence to obtain the named entity dataset.

4. The method for learning and training a tablet computer based on artificial intelligence according to claim 1, wherein: Perform relationship extraction based on the named entity dataset to obtain text feature semantic relationships, conduct knowledge fusion based on the text feature semantic relationships, and establish a learning material knowledge graph through the Neo4j graph database; The processing logic of the relationship extraction is as follows: generate graph nodes based on the named entity dataset, map each named entity in the named entity dataset to a node, generate a text node graph by connecting each node based on the text feature vector sequence, perform weighted calculation based on the graph attention mechanism through the gating unit, update the text node state to obtain an updated text node graph, and obtain the named entity relationship classification result through global graph pooling and classification processing on the updated text node graph; The named entity relationship classification results include: causal relationship, inclusion relationship, derivation relationship, opposition relationship, equivalence relationship, temporal relationship, and spatial relationship.

5. The method for learning and training of a tablet computer based on artificial intelligence according to claim 4, wherein: The processing logic of knowledge fusion is as follows: entity alignment processing is performed on the named entity dataset through the Word2Vec semantic similarity algorithm to obtain the aligned named entity dataset. Relationship fusion is carried out based on the named entity relationship classification results, and similar relationships are merged to obtain the named entity fusion relationship. The aligned named entity dataset and the named entity fusion relationship are uniformly processed according to the Neo4j data import format. Attribute labels and relationship labels are added to each named entity and corresponding relationship based on the aligned named entity dataset and the named entity fusion relationship to generate the knowledge graph of learning materials; The attribute labels include subject index, primary importance, secondary importance, primary difficulty, secondary difficulty, and tertiary difficulty; The relationship labels include causal relationship, inclusion relationship, derivative relationship, opposition relationship, equivalence relationship, temporal relationship, and spatial relationship.

6. The method for learning and training a tablet computer based on artificial intelligence according to claim 1, wherein: Identify the user's learning state and give message prompts to the user. Collect the facial micro-expression image data of the user through the front camera of the tablet computer. Analyze the facial micro-expression image data of the user through the second machine learning model to obtain the user learning state recognition result. The user learning state recognition result includes the focused state and the distracted state; Calculate the distraction frequency per unit time based on the user learning state recognition result, and perform threshold discrimination based on the distraction frequency. The threshold discrimination logic is: when the distraction frequency is greater than or equal to the first concentration threshold, the tablet computer executes the first message prompt. When the distraction frequency is greater than or equal to the second concentration threshold, the tablet computer executes the second message prompt; The first message prompt is: the tablet computer microphone plays the first prompt sound effect and vibrates, and at the same time, the screen display brightness changes uniformly and reciprocally between the original brightness and the maximum brightness. The duration of the first message prompt is 2 seconds; The second message prompt is: the tablet computer microphone plays the second prompt sound effect, stops the learning training, and pops up a rest notice.

7. A learning and training method for a tablet computer based on artificial intelligence according to claim 6, characterized in that: The second machine learning model is a pre-trained improved Yolov5 model. The logic of the second machine learning model for analyzing the user's facial micro-expression image is: at the input end, Mosaic is adopted for image data enhancement, and the facial micro-expression image data of four consecutive video frames are fused to obtain a fused image; Introduce the Focus module in the backbone network, perform adjacent domain downsampling processing on the fused image, and perform interval sampling and segmentation convolution processing on the fused image pixels to obtain preliminary image features; Introduce the BiFPN structure in the neck network, add weights to the input values of each channel in the preliminary image features respectively, and obtain the shallow fusion feature and the deep fusion feature through weighted calculation. Establish a cross-scale two-way connection between the convolutional layers, transfer the shallow fusion feature to the deep layer and add position information to the deep fusion feature, transfer the deep fusion feature to the shallow layer and add semantic information to the shallow fusion feature, and further perform feature fusion processing through the convolutional layer to obtain the weighted fusion image feature; Replace the detection head with a decoupled detection head in the head network. The classifier and regressor in parallel in the decoupled detection head are used to respectively perform bounding box prediction and class prediction on the weighted fusion image features, and the learning status recognition result is obtained by combining the non-maximum suppression algorithm; When training, the second machine learning model uses the DIoU loss function instead of the original CIoU loss function, and measures the loss by calculating the norm distance and overlapping area between the target box and the predicted box. The training set and validation set of the second machine learning model are obtained by dividing the preset facial micro-expression image set according to a ratio of 7:

3. The preset facial micro-expression image set is a learning status image with pre-labeled tags. The facial micro-expression image set includes focused images, distracted images, drowsy images, yawn images, and eye gaze deviation images. The labeled image tags include the focused state and the distracted state.

8. An artificial intelligence-based tablet learning and training system for implementing an artificial intelligence-based tablet learning and training method according to any one of claims 1-7, characterized in that, Including: A data storage module, a data processing module, a learning and training module, an interaction module, and a power supply module; The data storage module is used to store the learning material data set, the named entity data set, the learning material knowledge graph, and the user historical data. The data storage module is connected to the learning and training module and the data processing module, receives the learning material knowledge graph data sent by the data processing module, receives the user historical data sent by the learning and training module, receives the request instructions from the learning and training module and the data processing module, and sends the corresponding stored data to the learning and training module and the data processing module; The data processing module is used to generate the learning material knowledge graph corresponding to the learning material data set through the first machine learning model, identify the user's facial micro-expression image through the second machine learning model to obtain the user's learning status result, and calculate the message prompt data. The data processing module sends the learning material knowledge graph data and the user's learning status result data to the data storage module, and sends the message prompt data to the interaction module; The learning and training module is used to set the initial courses, derivative courses, derivative training question groups, and personalized training question groups, and provide courses and training questions for users. The learning and training module is connected to the data storage module, sends a request instruction to the data storage module, and receives the stored data corresponding to the instruction from the data storage module. The learning and training module is connected to the interaction module and receives the user's answer result data from the interaction module; The interaction module is used to perform human-computer interaction with the user, receive the user's learning status result sent by the data processing module and issue a message prompt, receive the courses and training questions of the learning and training module and display them to the user, record the user's answer result and send it to the learning and training module, receive the user's instructions and send them to the data storage module and the data processing module. The user's instructions include searching for named entities, selecting subjects, selecting courses, answering questions, and exiting the system; The power supply module is used to provide power for the data storage module, the data processing module, the learning and training module, and the interaction module.

Images