A learning machine control method, system and learning machine
Through deep learning technology, analyzing user's historical answer data and generating personalized exercises, solving the problem that traditional learning machines cannot meet students' personalized needs and improving learning results.
Patent Information
- Application Number
- CN202410278073.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-12
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2044-03-12
AI Technical Summary
Traditional learning machines have limited performance in personalized learning and cannot meet the unique learning needs and abilities of each student.
Using artificial intelligence analysis technology based on deep learning, we analyze user history answers and test papers to generate exercises related to weak knowledge points to meet users' personalized learning needs. The specific steps include obtaining user historical answer data, performing preliminary processing to obtain knowledge point keyword embedding vectors, performing feature encoding to obtain knowledge point semantic understanding feature vectors, and generating exercises based on these feature vectors.
Through personalized exercise generation, learning results are improved, students can improve their academic performance faster, and meet students' diversified learning needs.
Smart Images

Figure CN118313448B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of intelligent learning, and more specifically, to a learning machine control method, system and learning machine. Background Art
[0002] In 1993, with the initial breakthrough of technology, the first generation of learning machines emerged as a novel educational tool and quickly became a popular product among students. In that era, the main functions of learning machines focused on memorizing English words and reading Chinese texts, which provided great convenience for children's home learning. However, with the rapid development of technology and the renewal of educational concepts, traditional learning machines can no longer fully meet the diversified needs of modern students.
[0003] In the past thirty years, with the popularization of digital and network technology, the functions and forms of learning machines have undergone tremendous changes. Modern learning machines are no longer just a single learning tool, but a multimedia platform that integrates a variety of learning resources and functions. In addition to learning English words and Chinese texts, modern learning machines also cover learning content in multiple disciplines such as mathematics, physics, chemistry, and even more diversified learning fields such as programming and art.
[0004] However, despite the significant improvement in the functions and forms of learning machines, there are still some areas that can be improved. For example, traditional learning machines are limited in terms of personalized learning. Every student has different learning needs and abilities, but traditional learning machines can only provide uniform learning content and difficulty, which cannot meet the personalized needs of students.
[0005] Therefore, an optimized learning machine is needed. Summary of the invention
[0006] In order to solve the above technical problems, this application is proposed. The embodiments of this application provide a learning machine control method, system and learning machine, which adopts artificial intelligence analysis technology based on deep learning, analyzes the user's historical answer homework and test papers to obtain the user's knowledge point mastery, and generates exercises related to weak knowledge points based on the user's knowledge point mastery. In this way, the user's learning needs can be better met and the user's learning effect can be improved.
[0007] According to one aspect of the present application, a learning machine control method is provided, which includes:
[0008] Get the user's historical answer assignments and test papers;
[0009] Preliminarily processing the user's historical answer assignments and test papers to obtain multiple knowledge point keyword embedding vectors;
[0010] Performing feature encoding on the plurality of knowledge point keyword embedding vectors to obtain a knowledge point semantic understanding feature vector;
[0011] Based on the semantic understanding feature vector of the knowledge point, exercises related to weak knowledge points are generated.
[0012] In the above-mentioned learning machine control method, the user's historical answer homework and test papers are preliminarily processed to obtain multiple knowledge point keyword embedding vectors, including: preprocessing the user's historical answer homework and test papers to obtain the user's historical answer situation after preprocessing; segmenting the preprocessed user's historical answer situation and passing it through a word embedding layer to obtain the multiple knowledge point keyword embedding vectors.
[0013] In the above-mentioned learning machine control method, the feature encoding of the multiple knowledge point keyword embedding vectors to obtain the knowledge point semantic understanding feature vector includes: passing the multiple knowledge point keyword embedding vectors through a converter-based context encoder to obtain a first-scale knowledge point semantic understanding feature vector; arranging the multiple knowledge point keyword embedding vectors into a knowledge point keyword two-dimensional input matrix and then passing it through a text convolution neural network model based on a two-dimensional convolution kernel to obtain a second-scale knowledge point semantic understanding feature vector; performing probability density domain dimension-derived consistency projection on the first-scale knowledge point semantic understanding feature vector and the second-scale knowledge point semantic understanding feature vector to obtain the knowledge point semantic understanding feature vector.
[0014] In the above-mentioned learning machine control method, the multiple knowledge point keyword embedding vectors are passed through a converter-based context encoder to obtain a first-scale knowledge point semantic understanding feature vector, including: arranging the multiple knowledge point keyword embedding vectors into knowledge point keyword input vectors; converting the knowledge point keyword input vectors into knowledge point keyword query vectors and knowledge point keyword key vectors respectively through a learnable embedding matrix; calculating the product between the knowledge point keyword query vector and the transposed vector of the knowledge point keyword key vector to obtain a knowledge point keyword self-attention association matrix; normalizing the knowledge point keyword self-attention association matrix to obtain a knowledge point keyword standardized self-attention association matrix; inputting the knowledge point keyword standardized self-attention association matrix into a Softmax activation function for activation to obtain a knowledge point keyword self-attention feature matrix; multiplying the knowledge point keyword self-attention feature matrix with each knowledge point keyword embedding vector in the multiple knowledge point keyword embedding vectors as value vectors to obtain multiple knowledge point keyword context feature vectors; and cascading the multiple knowledge point keyword context feature vectors to obtain the first-scale knowledge point semantic understanding feature vector.
[0015] In the above-mentioned learning machine control method, the multiple knowledge point keyword embedding vectors are arranged into a knowledge point keyword two-dimensional input matrix and then passed through a text convolutional neural network model based on a two-dimensional convolution kernel to obtain a second-scale knowledge point semantic understanding feature vector, including: each layer of the text convolutional neural network model performs the following on the input data during the forward transmission process of the layer: convolution processing on the input data based on the two-dimensional convolution kernel to obtain a convolution feature map; mean pooling processing of each feature matrix along the channel dimension of the convolution feature map to obtain a pooled feature map; nonlinear activation of the pooled feature map to obtain an activation feature map; wherein the output of the last layer of the text convolutional neural network model is the second-scale knowledge point semantic understanding feature vector, and the input of the first layer of the text convolutional neural network model is the knowledge point keyword two-dimensional input matrix.
[0016] In the above-mentioned learning machine control method, the performing of probability density domain dimension-derived consistent projection on the first-scale knowledge point semantic understanding feature vector and the second-scale knowledge point semantic understanding feature vector to obtain the knowledge point semantic understanding feature vector comprises: performing probability density domain dimension-derived consistent projection on the first-scale knowledge point semantic understanding feature vector and the second-scale knowledge point semantic understanding feature vector according to the following calculation formula to obtain the knowledge point semantic understanding feature vector; wherein the following calculation formula is:
[0017]
[0018] in, represents the semantic understanding feature vector of the first scale knowledge point, represents the semantic understanding feature vector of the second scale knowledge point, represents calculating the distance between the semantic understanding feature vector of the first-scale knowledge point and the semantic understanding feature vector of the second-scale knowledge point, It means point multiplication by position. , and represents the weight hyperparameter, Represents the semantic understanding feature vector of the knowledge point.
[0019] In the above-mentioned learning machine control method, generating exercises related to weak knowledge points based on the knowledge point semantic understanding feature vector includes: passing the knowledge point semantic understanding feature vector through an exercise generator to obtain a generation result, and the generation result is an exercise related to weak knowledge points.
[0020] According to another aspect of the present application, there is provided a learning machine control system, comprising:
[0021] User-related data collection module, used to obtain the user's historical answer assignments and test papers;
[0022] A user-related data preliminary processing module, used for preliminary processing of the user's historical answer homework and test papers to obtain multiple knowledge point keyword embedding vectors;
[0023] A knowledge point semantic understanding feature generation module, used for performing feature encoding on the plurality of knowledge point keyword embedding vectors to obtain a knowledge point semantic understanding feature vector;
[0024] The knowledge point related exercise result generation module is used to generate weak knowledge point related exercises based on the knowledge point semantic understanding feature vector.
[0025] According to another aspect of the present application, a learning machine is provided, comprising a memory and a processor coupled to the memory, wherein the processor is configured to execute the learning machine control method as described in any one of the above items based on instructions stored in the memory.
[0026] Compared with the prior art, the learning machine control method, system and learning machine provided by the present application adopt artificial intelligence analysis technology based on deep learning, analyze the user's historical answer homework and test papers to obtain the user's knowledge mastery, and generate exercises related to weak knowledge points based on the user's knowledge mastery. In this way, the user's learning needs can be better met and the user's learning effect can be improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.
[0028] Figure 1 Flow chart of a learning machine control method according to an embodiment of the present application.
[0029] Figure 2 The present invention is a flowchart of performing preliminary processing on the user's historical answer assignments and test papers to obtain multiple knowledge point keyword embedding vectors in the learning machine control method according to an embodiment of the present application.
[0030] Figure 3 The present invention is a flowchart of performing feature encoding on the multiple knowledge point keyword embedding vectors to obtain the knowledge point semantic understanding feature vectors in the learning machine control method according to an embodiment of the present application.
[0031] Figure 4The present invention is a flowchart of a learning machine control method according to an embodiment of the present application, in which the multiple knowledge point keyword embedding vectors are passed through a converter-based context encoder to obtain a first-scale knowledge point semantic understanding feature vector.
[0032] Figure 5 4 is a system block diagram of a learning machine control system according to an embodiment of the present application.
[0033] Figure 6 is a block diagram of a learning machine according to an embodiment of the present application. DETAILED DESCRIPTION
[0034] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.
[0035] As mentioned in the background technology above, since the birth of the first generation of learning machines in 1993, learning machines have undergone tremendous changes. Initially, they mainly helped students memorize English words and read texts, but now they have evolved into a multimedia platform that integrates a variety of learning resources and functions, covering multiple subject areas. However, despite the improvements in the functions and forms of learning machines, they still need to be improved to meet the needs of students' personalized learning. Therefore, an optimized learning machine is expected.
[0036] In recent years, deep learning and neural networks have been widely used in computer vision, natural language processing, text signal processing and other fields. In addition, deep learning and neural networks have also shown a level close to or even beyond that of humans in image classification, object detection, semantic segmentation, text translation and other fields. The development of deep learning and neural networks provides new solutions and solutions for learning machine control.
[0037] Figure 1 Flowchart of the learning machine control method according to the embodiment of the present application. Figure 1 As shown, the learning machine control method according to the embodiment of the present application includes: S110, obtaining the user's historical answering homework and test papers; S120, preliminarily processing the user's historical answering homework and test papers to obtain multiple knowledge point keyword embedding vectors; S130, feature encoding the multiple knowledge point keyword embedding vectors to obtain a knowledge point semantic understanding feature vector; S140, generating weak knowledge point related exercises based on the knowledge point semantic understanding feature vector.
[0038] Specifically, in the technical solution of the present application, first, the user's historical answer homework and test papers are obtained. It should be well understood that through the user's historical answer homework and test papers, the learning machine can understand the learning characteristics and level of each student, and tailor learning plans and exercises for each student. That is, the learning machine can identify which knowledge points the students have difficulties or errors in by analyzing historical answer homework and test papers, so as to provide relevant exercises and guidance in a targeted manner to help students overcome their weak links. Based on the analysis of historical answer data, the learning machine can adjust the learning path and content to make the learning process more efficient and more targeted, helping students improve their academic performance faster.
[0039] In one embodiment of the present application, a possible implementation method for obtaining a user's historical answer homework and test papers can be: the learning machine is generally bound to a personal account when used, and when the user completes the homework and test papers on the learning machine, the relevant data will be recorded for subsequent analysis and use by the learning machine. Current learning machines also allow users to manually import historical answer homework and test paper data. It is worth noting that obtaining the user's relevant data needs to comply with relevant privacy policies and regulations to protect the user's personal information from being leaked.
[0040] Figure 2 The flowchart of the learning machine control method according to the embodiment of the present application is to perform preliminary processing on the user's historical answer homework and test papers to obtain multiple knowledge point keyword embedding vectors. Figure 2 As shown, the feature encoding of the multiple knowledge point keyword embedding vectors to obtain the knowledge point semantic understanding feature vectors includes: S121, preprocessing the user's historical answer homework and test papers to obtain the user's historical answer situation after preprocessing; S122, segmenting the preprocessed user's historical answer situation and passing it through a word embedding layer to obtain the multiple knowledge point keyword embedding vectors.
[0041] In order to accurately and efficiently extract key information and help the learning machine better understand the students' learning situation and needs, it is necessary to clean and standardize the original data to make it more suitable for subsequent analysis and processing, that is, in the technical solution of this application, the user's historical answer assignments and test papers are preprocessed to obtain the user's historical answer situation after preprocessing. By preprocessing the original data, the data quality can be improved, interference can be reduced, and subsequent analysis can be made more accurate and reliable. It provides more accurate and useful data for the learning machine analysis, which can speed up the learning machine's analysis and understanding of the user's historical answer situation.
[0042] In an embodiment of the present application, an implementable manner for preprocessing the user's historical answering assignments and test papers to obtain the preprocessed user's historical answering situation may be as follows: Special characters and punctuation marks usually do not contain actual information. By using regular expressions, special characters and punctuation marks in the text can be removed, which helps to simplify the text content. Stop words are common words that have no actual meaning in text analysis, such as "de" (的), "shi" (是), etc. They can be filtered through a stop word list, which can make keywords more prominent and accurately reflect the theme and content of the text.
[0043] Subsequently, the preprocessed user's historical answering situation is segmented and then passed through a word embedding layer to obtain multiple knowledge point keyword embedding vectors. It should be well understood that the word embedding layer is a commonly used layer in deep learning models, which is used to convert discrete word representations into continuous word vector representations. By mapping words into a continuous vector space, word embedding can capture the semantic relationships between words, making words with similar meanings closer in the vector space, thereby improving the understanding of the semantic relationships between knowledge point keywords. Converting text data into embedding vectors can reduce the high-dimensional text data to a low-dimensional continuous vector space, improving the data processing efficiency and the model training speed. The embedding vectors can capture the semantic relationships between words, making words with similar semantics closer in the vector space, which is beneficial for the model to better understand and process text data.
[0044] Figure 3 The following is a flowchart for feature encoding the multiple knowledge point keyword embedding vectors to obtain a knowledge point semantic understanding feature vector in the learning machine control method according to an embodiment of the present application. As Figure 3 shown, the feature encoding of the multiple knowledge point keyword embedding vectors to obtain a knowledge point semantic understanding feature vector includes: S131, passing the multiple knowledge point keyword embedding vectors through a context encoder based on a Transformer to obtain a first-scale knowledge point semantic understanding feature vector; S132, arranging the multiple knowledge point keyword embedding vectors into a two-dimensional input matrix of knowledge point keywords and then passing it through a text convolutional neural network model based on a two-dimensional convolutional kernel to obtain a second-scale knowledge point semantic understanding feature vector; S133, performing a probability density domain dimension derivative consistency projection on the first-scale knowledge point semantic understanding feature vector and the second-scale knowledge point semantic understanding feature vector to obtain the knowledge point semantic understanding feature vector.
[0045] In order to enable the learning machine to better understand and represent the semantic relationship between knowledge points and improve its ability to understand and analyze students' learning situations, in the technical solution of the present application, the multiple knowledge point keyword embedding vectors are passed through a converter-based context encoder to obtain a first-scale knowledge point semantic understanding feature vector. The converter-based context encoder can effectively integrate the context information between multiple knowledge point keyword embedding vectors, helping the learning machine to better understand the semantic relationship and connection between knowledge points. Through the context encoder, the semantic information of each knowledge point keyword can be deeply understood and represented, improving the learning machine's ability to grasp the semantic association between knowledge points.
[0046] Figure 4 The flowchart of the learning machine control method according to the embodiment of the present application is to pass the multiple knowledge point keyword embedding vectors through the context encoder based on the converter to obtain the first scale knowledge point semantic understanding feature vector. Figure 4 As shown, the plurality of knowledge point keyword embedding vectors are passed through a converter-based context encoder to obtain a first-scale knowledge point semantic understanding feature vector, including: S1311, arranging the plurality of knowledge point keyword embedding vectors into a knowledge point keyword input vector; S1312, converting the knowledge point keyword input vector into a knowledge point keyword query vector and a knowledge point keyword key vector respectively through a learnable embedding matrix; S1313, calculating the product between the knowledge point keyword query vector and the transposed vector of the knowledge point keyword key vector to obtain a knowledge point keyword self-attention association matrix; S1314, The knowledge point keyword self-attention feature matrix is normalized to obtain a knowledge point keyword standardized self-attention association matrix; S1315, the knowledge point keyword standardized self-attention association matrix is input into the Softmax activation function for activation to obtain a knowledge point keyword self-attention feature matrix; S1316, the knowledge point keyword self-attention feature matrix is multiplied with each knowledge point keyword embedding vector in the multiple knowledge point keyword embedding vectors as a value vector to obtain multiple knowledge point keyword context feature vectors; S1317, the multiple knowledge point keyword context feature vectors are cascaded to obtain the first-scale knowledge point semantic understanding feature vector.
[0047] In order to capture the local semantic relationship and pattern between knowledge point keywords, in the technical solution of the present application, the multiple knowledge point keyword embedding vectors are arranged into a knowledge point keyword two-dimensional input matrix and then a text convolutional neural network model based on a two-dimensional convolution kernel is used to obtain a second-scale knowledge point semantic understanding feature vector. The text convolutional neural network model based on a two-dimensional convolutional kernel can effectively extract local features from the knowledge point keyword two-dimensional input matrix, and the convolutional neural network model can effectively reduce the amount of model parameters and improve the training efficiency and generalization ability of the model by sharing parameters. It is also worth mentioning that the convolutional neural network model has the characteristics of translation invariance, which can identify and capture the semantic features of knowledge point keywords at different positions, and improve the globality of the model's semantic understanding of knowledge points.
[0048] In an embodiment of the present application, after arranging the multiple knowledge point keyword embedding vectors into a knowledge point keyword two-dimensional input matrix, a possible implementation method for obtaining a second-scale knowledge point semantic understanding feature vector through a text convolutional neural network model based on a two-dimensional convolutional kernel may be as follows: each layer of the text convolutional neural network model based on a two-dimensional convolutional kernel performs the following on the input data during the forward transfer process of the layer: performing convolution processing on the input data based on the two-dimensional convolutional kernel to obtain a convolution feature map; performing mean pooling processing on the convolution feature map of each feature matrix along the channel dimension to obtain a pooled feature map; performing nonlinear activation on the pooled feature map to obtain an activation feature map; wherein the output of the last layer of the text convolutional neural network model is the second-scale knowledge point semantic understanding feature vector, and the input of the first layer of the text convolutional neural network model is the knowledge point keyword two-dimensional input matrix.
[0049] In the technical solution of the present application, it is considered that the different processing methods of the semantic understanding feature vector of the first scale knowledge point and the semantic understanding feature vector of the second scale knowledge point in feature extraction and representation will lead to the difference in the data manifolds in which they are located. Specifically, the semantic understanding feature vector of the first scale knowledge point is obtained by a context encoder based on a converter, which mainly relies on word embedding and context encoding, and can capture semantic information at the word or phrase level. Since the converter model has good context understanding ability when processing text, the first scale feature vector pays more attention to the semantic association and context information between words, so that a data manifold of semantic relevance is formed in a high-dimensional space. The semantic understanding feature vector of the second scale knowledge point is obtained by a text convolutional neural network model based on a two-dimensional convolution kernel, which mainly extracts features in a two-dimensional space through convolution operations and pooling operations, and can capture a wider range of semantic information. The text convolutional neural network extracts local features by sliding the convolution kernel on a two-dimensional matrix, and reduces the dimension through pooling operations, so that a data manifold based on local and global semantic information is formed in a high-dimensional space. Therefore, the first-scale knowledge point semantic understanding feature vector and the second-scale knowledge point semantic understanding feature vector correspond to different high-dimensional data manifolds respectively. Since the first-scale knowledge point semantic understanding feature vector and the second-scale knowledge point semantic understanding feature vector correspond to different high-dimensional data manifolds respectively, if the first-scale knowledge point semantic understanding feature vector and the second-scale knowledge point semantic understanding feature vector are directly fused in a cascade or position-weighted sum manner, this will not only cause the fused knowledge point semantic understanding feature vector to have blurred boundaries in the class boundary area, but also reduce the correlation between the first-scale knowledge point semantic understanding feature vector and the second-scale knowledge point semantic understanding feature vector, thereby affecting the accuracy of the generation of weak knowledge point related exercises. Based on this, in the technical solution of the present application, the first-scale knowledge point semantic understanding feature vector and the second-scale knowledge point semantic understanding feature vector are subjected to probability density domain dimension-derived consistency projection.
[0050] Specifically, the first-scale knowledge point semantic understanding feature vector and the second-scale knowledge point semantic understanding feature vector are subjected to probability density domain dimension-derived consistency projection to obtain the knowledge point semantic understanding feature vector according to the following calculation formula; wherein the following calculation formula is:
[0051]
[0052] in, represents the semantic understanding feature vector of the first scale knowledge point, represents the semantic understanding feature vector of the second scale knowledge point, represents calculating the distance between the semantic understanding feature vector of the first-scale knowledge point and the semantic understanding feature vector of the second-scale knowledge point, It means point multiplication by position. , and represents the weight hyperparameter, Represents the semantic understanding feature vector of the knowledge point.
[0053] Here, in the technical solution of the present application, the first-scale knowledge point semantic understanding feature vector and the second-scale knowledge point semantic understanding feature vector are subjected to probability density domain dimension-derived consistency projection, which utilizes the idea of probability density domain dimension-derived to map the first-scale knowledge point semantic understanding feature vector and the second-scale knowledge point semantic understanding feature vector into a mutually related probability density domain space, so that in this space, the distribution of the first-scale knowledge point semantic understanding feature vector and the second-scale knowledge point semantic understanding feature vector are consistent, that is, the probability density functions of the two are the same or similar. Furthermore, utilizing the idea of consistent projection, the data in the probability density domain space is projected into the same classification feature space, so that in this space, the category information of the first-scale knowledge point semantic understanding feature vector and the second-scale knowledge point semantic understanding feature vector is retained and enhanced, that is, the category boundaries of the two are clearer and more separated, so as to improve the accuracy of the knowledge point semantic understanding feature vector in generating weak knowledge point related exercises through the exercise generator.
[0054] Finally, the knowledge point semantic understanding feature vector is passed through an exercise generator to obtain a generated result, and the generated result is exercises related to weak knowledge points. In the technical solution of the present application, the essence of the exercise generator is a generator based on a generative adversarial network. The exercise generator can generate exercises related to weak knowledge points based on the knowledge point semantic understanding feature vector, which can help users to make up for the deficiencies in knowledge points in a targeted manner and provide personalized learning support. This customized exercise can better meet the user's learning needs and improve learning effects. The generated exercises can help users better understand their learning status, timely discover and make up for weak knowledge points, provide personalized feedback and guidance, and promote the user's learning progress.
[0055] In an embodiment of the present application, the knowledge point semantic understanding feature vector is passed through an exercise generator to obtain a generation result, and the generation result is a weak knowledge point related exercise. One possible implementation method may be: the exercise generator includes a discriminator and a generator, wherein the generator is used to generate weak knowledge point related exercises, the discriminator is used to calculate the difference between the generated weak knowledge point related exercises and the real weak knowledge point related exercises, and the network parameters of the generator are updated through the directional propagation algorithm of gradient descent to obtain a generator with the function of generating exercises, that is, the exercise generator.
[0056] In summary, the learning machine control method based on the embodiment of the present application is explained, which adopts artificial intelligence analysis technology based on deep learning, analyzes the user's historical answer homework and test papers to obtain the user's knowledge point mastery, and generates exercises related to weak knowledge points based on the user's knowledge point mastery. In this way, the user's learning needs can be better met and the user's learning effect can be improved.
[0057] Figure 5 : is a system block diagram of a learning machine control system according to an embodiment of the present application. Figure 5 As shown, according to the learning machine control system 100 of the embodiment of the present application, it includes: a user-related data acquisition module 110, which is used to obtain the user's historical answer homework and test papers; a user-related data preliminary processing module 120, which is used to perform preliminary processing on the user's historical answer homework and test papers to obtain multiple knowledge point keyword embedding vectors; a knowledge point semantic understanding feature generation module 130, which is used to feature encode the multiple knowledge point keyword embedding vectors to obtain a knowledge point semantic understanding feature vector; a knowledge point related exercise result generation module 140, which is used to generate weak knowledge point related exercises based on the knowledge point semantic understanding feature vector.
[0058] Here, those skilled in the art will appreciate that the specific functions and operations of the various units and modules in the above-mentioned learning machine control system 100 have been described in detail above. Figures 1 to 4 The description of the learning machine control method has been described in detail, and therefore, its repeated description will be omitted.
[0059] In summary, the learning machine control system 100 according to the embodiment of the present application is explained, which adopts artificial intelligence analysis technology based on deep learning, analyzes the user's historical answering homework and test papers to obtain the user's knowledge point mastery, and generates exercises related to weak knowledge points based on the user's knowledge point mastery. In this way, the user's learning needs can be better met and the user's learning effect can be improved.
[0060] As described above, the learning machine control system 100 according to the embodiment of the present application can be implemented in various wireless terminals, such as a server for learning machine control. In one example, the learning machine control system 100 according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the learning machine control system 100 can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the learning machine control system 100 can also be one of the many hardware modules of the wireless terminal.
[0061] Alternatively, in another example, the learning machine control system 100 and the wireless terminal may also be separate devices, and the learning machine control system 100 may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.
[0062] Below, reference Figure 6 To describe the learning machine according to an embodiment of the present application. Figure 6 is a block diagram of a learning machine according to an embodiment of the present application. Figure 6 As shown, the learning machine 10 includes one or more processors 11 and a memory 12 .
[0063] The processor 11 may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the learning machine 10 to perform desired functions. The memory 12 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, a random access memory (RAM) and / or a cache memory (cache), etc. The non-volatile memory may include, for example, a read-only memory (ROM), a hard disk, a flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 11 may run the program instructions to implement the functions of the various embodiments of the present application described above. Various contents such as user historical answer assignments and test papers may also be stored in the computer-readable storage medium.
[0064] In one example, the learning machine 10 may further include: an input device 13 and an output device 14, which are interconnected via a bus system and / or other forms of connection mechanisms (not shown). The input device 13 may include, for example, a keyboard, a mouse, etc. The output device 14 may output various information to the outside, including exercises related to weak knowledge points, etc. The output device 14 may include, for example, a display, a speaker, a printer, a communication network and a remote output device connected thereto, etc.
[0065] In the several embodiments provided in this application, it should be understood that the disclosed method, system or device can be implemented in other ways. For example, the system embodiment described above is only illustrative, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.
[0066] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0067] In addition, each functional module in each embodiment of the present application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of hardware plus software functional modules.
[0068] It is obvious to those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and that the present application can be implemented in other specific forms without departing from the spirit or essential features of the present application.
[0069] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present application is limited by the appended claims rather than the above description, so it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present application. Any attached figure mark in the claims should not be regarded as limiting the claims involved.
[0070] In addition, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the system claim can also be implemented by one unit or device through software or hardware. The second and other words are used to indicate names, but not to indicate any particular order.
[0071] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present application and are not intended to limit it. Although the present application has been described in detail with reference to the preferred embodiments, a person of ordinary skill in the art should understand that the technical solution of the present application may be modified or replaced by equivalents without departing from the spirit of the technical solution of the present application.
Claims
1. A learning machine control method, characterized in that: include: Get the user's historical answer assignments and test papers; Preliminarily processing the user's historical answer assignments and test papers to obtain multiple knowledge point keyword embedding vectors; Performing feature encoding on the plurality of knowledge point keyword embedding vectors to obtain a knowledge point semantic understanding feature vector; Generate exercises related to weak knowledge points based on the semantic understanding feature vector of the knowledge points; The step of performing feature encoding on the plurality of knowledge point keyword embedding vectors to obtain a knowledge point semantic understanding feature vector includes: Passing the plurality of knowledge point keyword embedding vectors through a transformer-based context encoder to obtain a first-scale knowledge point semantic understanding feature vector; Arranging the plurality of knowledge point keyword embedding vectors into a knowledge point keyword two-dimensional input matrix and then passing the matrix through a text convolutional neural network model based on a two-dimensional convolution kernel to obtain a second-scale knowledge point semantic understanding feature vector; Performing probability density domain dimension-derived consistency projection on the first-scale knowledge point semantic understanding feature vector and the second-scale knowledge point semantic understanding feature vector to obtain the knowledge point semantic understanding feature vector; Among them, performing probability density domain dimension-derived consistency projection on the first-scale knowledge point semantic understanding feature vector and the second-scale knowledge point semantic understanding feature vector to obtain the knowledge point semantic understanding feature vector, including: performing probability density domain dimension-derived consistency projection on the first-scale knowledge point semantic understanding feature vector and the second-scale knowledge point semantic understanding feature vector according to the following calculation formula to obtain the knowledge point semantic understanding feature vector; The calculation formula is as follows: Among them, V1 represents the semantic understanding feature vector of the first-scale knowledge point, V2 represents the semantic understanding feature vector of the second-scale knowledge point, d(V1, V2) represents the calculation of the distance between the semantic understanding feature vector of the first-scale knowledge point and the semantic understanding feature vector of the second-scale knowledge point, ⊙ represents the point multiplication by position, α, β and γ represent weight hyperparameters, and w represents the semantic understanding feature vector of the knowledge point.
2. The learning machine control method according to claim 1, characterized in that: The user's historical answer assignments and test papers are preliminarily processed to obtain multiple knowledge point keyword embedding vectors, including: Preprocessing the user's historical answer assignments and test papers to obtain the user's historical answer status after preprocessing; The pre-processed user historical answers are segmented and passed through a word embedding layer to obtain the multiple knowledge point keyword embedding vectors.
3. The learning machine control method according to claim 2, characterized in that: The plurality of knowledge point keyword embedding vectors are passed through a converter-based context encoder to obtain a first-scale knowledge point semantic understanding feature vector, including: Arranging the plurality of knowledge point keyword embedding vectors into a knowledge point keyword input vector; The knowledge point keyword input vector is converted into a knowledge point keyword query vector and a knowledge point keyword key vector respectively through a learnable embedding matrix; Calculating the product between the knowledge point keyword query vector and the transposed vector of the knowledge point keyword key vector to obtain a knowledge point keyword self-attention association matrix; Normalizing the knowledge point keyword self-attention association matrix to obtain a knowledge point keyword standardized self-attention association matrix; Input the knowledge point keyword standardized self-attention association matrix into the Softmax activation function for activation to obtain the knowledge point keyword self-attention feature matrix; Multiplying the knowledge point keyword self-attention feature matrix and each knowledge point keyword embedding vector in the multiple knowledge point keyword embedding vectors as value vectors to obtain multiple knowledge point keyword context feature vectors; The multiple knowledge point keyword context feature vectors are cascaded to obtain the first-scale knowledge point semantic understanding feature vector.
4. The learning machine control method according to claim 3, characterized in that: Arranging the plurality of knowledge point keyword embedding vectors into a knowledge point keyword two-dimensional input matrix and then passing through a text convolutional neural network model based on a two-dimensional convolution kernel to obtain a second-scale knowledge point semantic understanding feature vector, including: Each layer of the text convolutional neural network model performs the following operations on the input data during the forward transmission of the layer: Performing convolution processing on the input data based on the two-dimensional convolution kernel to obtain a convolution feature map; Performing mean pooling processing on each feature matrix along the channel dimension on the convolution feature map to obtain a pooled feature map; Performing nonlinear activation on the pooled feature map to obtain an activated feature map; Among them, the output of the last layer of the text convolutional neural network model is the second-scale knowledge point semantic understanding feature vector, and the input of the first layer of the text convolutional neural network model is the knowledge point keyword two-dimensional input matrix.
5. The learning machine control method according to claim 4, characterized in that: Based on the knowledge point semantic understanding feature vector, generating exercises related to weak knowledge points, including: passing the knowledge point semantic understanding feature vector through an exercise generator to obtain a generation result, wherein the generation result is an exercise related to weak knowledge points.
6. A learning machine control system, characterized in that: include: User-related data collection module, used to obtain the user's historical answer assignments and test papers; A user-related data preliminary processing module, used for preliminary processing of the user's historical answer homework and test papers to obtain multiple knowledge point keyword embedding vectors; A knowledge point semantic understanding feature generation module, used for performing feature encoding on the plurality of knowledge point keyword embedding vectors to obtain a knowledge point semantic understanding feature vector; A knowledge point related exercise result generation module, used for generating weak knowledge point related exercises based on the knowledge point semantic understanding feature vector; The step of performing feature encoding on the plurality of knowledge point keyword embedding vectors to obtain a knowledge point semantic understanding feature vector includes: Passing the plurality of knowledge point keyword embedding vectors through a transformer-based context encoder to obtain a first-scale knowledge point semantic understanding feature vector; Arranging the plurality of knowledge point keyword embedding vectors into a knowledge point keyword two-dimensional input matrix and then passing the matrix through a text convolutional neural network model based on a two-dimensional convolution kernel to obtain a second-scale knowledge point semantic understanding feature vector; Performing probability density domain dimension-derived consistency projection on the first-scale knowledge point semantic understanding feature vector and the second-scale knowledge point semantic understanding feature vector to obtain the knowledge point semantic understanding feature vector; Among them, performing probability density domain dimension-derived consistency projection on the first-scale knowledge point semantic understanding feature vector and the second-scale knowledge point semantic understanding feature vector to obtain the knowledge point semantic understanding feature vector, including: performing probability density domain dimension-derived consistency projection on the first-scale knowledge point semantic understanding feature vector and the second-scale knowledge point semantic understanding feature vector according to the following calculation formula to obtain the knowledge point semantic understanding feature vector; The calculation formula is as follows: Among them, V1 represents the semantic understanding feature vector of the first-scale knowledge point, V2 represents the semantic understanding feature vector of the second-scale knowledge point, d(V1, V2) represents the calculation of the distance between the semantic understanding feature vector of the first-scale knowledge point and the semantic understanding feature vector of the second-scale knowledge point, ⊙ represents the point multiplication by position, α, β and γ represent weight hyperparameters, and w represents the semantic understanding feature vector of the knowledge point.
7. The learning machine control system according to claim 6, characterized in that: The user-related data preliminary processing module includes: A data preprocessing unit, used to preprocess the user's historical answer assignments and test papers to obtain the user's historical answer status after preprocessing; The keyword embedding unit is used to segment the pre-processed user historical answers and pass them through a word embedding layer to obtain the keyword embedding vectors of the multiple knowledge points.
8. A learning machine comprising a memory and a processor coupled to the memory, characterized in that: The processor is configured to execute the learning machine control method according to any one of claims 1 to 5 based on the instructions stored in the memory.
Citation Information
Patent Citations
Weak knowledge point identification method and learning path planning method and device
CN114254208A
Smart learning knowledge point mining method and system based on big data
CN116541538A
Multi-label identification method and apparatus, electronic device, and storage medium
WO2023134084A1
