A knowledge point prediction method, device, equipment and storage medium
By using predefined keywords and exercise text matching to generate weak labels, and training classifiers with text features and question type features, the problem of inefficient prediction of knowledge points in the existing technology is solved, and the prediction accuracy and reliability are improved. It is suitable for Personalized learning programs in cognitive diagnosis.
Patent Information
- Application Number
- CN202411214791.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-02
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-09-02
AI Technical Summary
The existing machine learning-based knowledge point prediction methods are inefficient and cannot effectively predict students' mastery of course knowledge points, resulting in poor cognitive diagnosis.
By matching predefined keywords with exercise text samples, weak labels are generated, and the exercise text is converted into high-dimensional vectors, text features and question type features are extracted based on context information, and the classifier is trained to predict knowledge points after fusion.
It improves the accuracy and reliability of knowledge point prediction, enhances the training effect of classifiers, reduces dependence on manual annotation, and improves learning efficiency and quality.
Smart Images

Figure CN118733745B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of natural language processing technology, and in particular to a knowledge point prediction method, device, equipment and storage medium. Background Art
[0002] In recent years, with the deep integration of education and information technology, cognitive diagnosis as an intelligent diagnostic analysis technology has gradually become a research hotspot. This study aims to accurately reveal students' mastery of various knowledge points in the course and provide a scientific basis for the formulation of personalized learning plans and teaching strategies. Cognitive diagnosis is the premise and foundation for personalized learning. The prediction of exercise-related knowledge points is an important part of cognitive diagnosis research. It provides support for predicting students' knowledge proficiency and improving their knowledge application ability by analyzing the relationship between exercises and knowledge points, thereby helping students to strengthen their knowledge in a targeted manner and improve learning efficiency and quality.
[0003] Among the current mainstream methods, there are two main methods for predicting the knowledge points associated with exercises: manual annotation and machine learning. Manual annotation is time-consuming, labor-intensive, and inefficient. This has the problems of low efficiency and omissions in judgment, which will have a negative impact on cognitive diagnosis. Therefore, methods based on machine learning have gradually become a research hotspot. In recent years, automatic prediction methods based on text classification have continued to emerge, such as methods that use TF-IDF to extract features and support vector machines to annotate knowledge points, methods based on vector space models, and methods that use convolutional neural networks and teaching and research knowledge enhancement. However, these methods usually only consider the one-sided features of the text, resulting in the knowledge point prediction results that cannot achieve the expected results. Summary of the invention
[0004] The purpose of the present invention is to solve the problem that the prediction results of knowledge points based on current machine learning methods cannot achieve the expected effect. The present invention provides a knowledge point prediction method, device, equipment and storage medium.
[0005] The technical solution of the embodiment of the present application is implemented as follows:
[0006] A first aspect of an embodiment of the present application provides a knowledge point prediction method, comprising:
[0007] Using predefined keywords to match the exercise text samples, determining the weak labels corresponding to each of the exercise text samples; the keywords represent different knowledge categories;
[0008] Convert each component element in the exercise text sample into a high-dimensional vector, combine the context information of the exercise text sample to obtain text features, and obtain question type features based on the feature elements in the exercise text sample;
[0009] A feature vector is obtained based on the fusion of the text feature and the question type feature, and a preset classifier is trained using the feature vector and the weak label to obtain a target classifier;
[0010] The exercise text to be predicted is input into the target classifier to determine the predicted knowledge points corresponding to the exercise text to be predicted.
[0011] Optionally, converting each component element in the exercise text sample into a high-dimensional vector and combining the context information of the exercise text sample to obtain text features includes:
[0012] Using the pre-trained model Bert as a feature extractor, each component element in the exercise text sample is converted into a high-dimensional vector;
[0013] Inputting the high-dimensional vector into a bidirectional long short-term memory network and fusing the context information of the exercise text sample to obtain a feature vector of each of the component elements;
[0014] The sum of the feature vectors of the component elements is determined as the text feature of the exercise text sample.
[0015] Optionally, obtaining question type features based on characteristic elements in the exercise text sample includes:
[0016] Performing word segmentation processing on the exercise text sample to obtain multiple feature elements;
[0017] Based on the correlation between the different characteristic elements, the question type characteristics are obtained.
[0018] Optionally, the step of obtaining a feature vector based on the fusion of the text feature and the question type feature, and training a preset classifier using the feature vector and the weak label to obtain a target classifier includes:
[0019] Input the feature vector and the weak label into a preset classifier, and output a mapping relationship between the feature vector and the weak label;
[0020] Adjusting the parameters of the preset classifier based on the loss function corresponding to the mapping relationship until the loss function reaches a preset threshold;
[0021] The preset classifier when the loss function reaches a preset threshold is determined as the target classifier.
[0022] Optionally, the step of inputting the to-be-predicted exercise text into the target classifier and determining the predicted knowledge points corresponding to the to-be-predicted exercise text includes:
[0023] Inputting the to-be-predicted exercise text into the target classifier to obtain at least one group of candidate knowledge points and the prediction probability of each group of the knowledge points;
[0024] A group of candidate knowledge points with the largest prediction probability is determined as the predicted knowledge points corresponding to the exercise text to be predicted.
[0025] The second aspect of the embodiment of the present application provides a knowledge point prediction device, including: a weak label generation module, a feature extraction module, a training module and a determination module, wherein:
[0026] The weak label generation module is configured to use predefined keywords to match the exercise text samples to determine the weak labels corresponding to each of the exercise text samples; the keywords represent different knowledge categories;
[0027] The feature extraction module is configured to convert each component element in the exercise text sample into a high-dimensional vector, obtain text features in combination with context information of the exercise text sample, and obtain question type features based on the feature elements in the exercise text sample;
[0028] The training module is configured to obtain a feature vector based on the fusion of the text feature and the question type feature, and train a preset classifier using the feature vector and the weak label to obtain a target classifier;
[0029] The determination module is configured to input the to-be-predicted exercise text into the target classifier to determine the predicted knowledge points corresponding to the to-be-predicted exercise text.
[0030] Optionally, the feature extraction module is specifically configured as follows:
[0031] Using the pre-trained model Bert as a feature extractor, each component element in the exercise text sample is converted into a high-dimensional vector;
[0032] Inputting the high-dimensional vector into a bidirectional long short-term memory network and fusing the context information of the exercise text sample to obtain a feature vector of each of the component elements;
[0033] The sum of the feature vectors of the component elements is determined as the text feature of the exercise text sample.
[0034] Optionally, the feature extraction module is specifically configured as follows:
[0035] Performing word segmentation processing on the exercise text sample to obtain multiple feature elements;
[0036] Based on the different characteristic elements and the association relationship between the characteristic elements, the question type characteristics are obtained.
[0037] A third aspect of an embodiment of the present application provides an electronic device, comprising a processor and a memory; the memory stores a computer program, wherein the computer program implements the knowledge point prediction method described in the first aspect when executed by the processor.
[0038] A fourth aspect of an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.
[0039] Compared with the prior art, the technical solution provided by this application has the following beneficial effects:
[0040] The present invention provides a knowledge point prediction method, device, equipment and storage medium, which uses a predefined keyword list to match with an exercise text sample, determines the weak label corresponding to each of the exercise text samples, converts each component element in the exercise text sample into a high-dimensional vector, combines the context information of the exercise text sample, obtains text features, and obtains question type features based on the feature elements in the exercise text sample, obtains a feature vector based on the fusion of the text feature and the question type feature, and uses the feature vector and the weak label to train a preset classifier to obtain a target classifier, inputs the exercise text to be predicted into the target classifier, and determines the predicted knowledge point corresponding to the exercise text to be predicted. The present invention trains the classifier using the weak label, text feature and question type feature of the exercise text, fully considering the feature type of the exercise text, thereby improving the training effect of the classifier, and further improving the accuracy and reliability of the predicted knowledge points. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 A schematic diagram of a flow chart of a knowledge point prediction method provided in an embodiment of the present application;
[0042] Figure 2 A schematic diagram of the structure of a knowledge point prediction device provided in an embodiment of the present application;
[0043] Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0044] Below, embodiments of the present application will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present application. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present application. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present application.
[0045] The terms used herein are only for describing specific embodiments and are not intended to limit the present application. The terms "include", "comprising", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.
[0046] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.
[0047] Some block diagrams and / or flow charts are shown in the accompanying drawings. It should be understood that some blocks or combinations thereof in the block diagrams and / or flow charts may be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that these instructions, when executed by the processor, may create a device for implementing the functions / operations described in these block diagrams and / or flow charts.
[0048] In some embodiments, see Figure 1 , Figure 1 A schematic diagram of a flow chart of a knowledge point prediction method provided in an embodiment of the present application; the knowledge point prediction method provided in an embodiment of the present application includes:
[0049] S110, using predefined keywords to match the exercise text samples to determine the weak labels corresponding to each exercise text sample; the keywords represent different knowledge categories.
[0050] Here, a fine-tuning method based on LoRA (Low-Rank Adaptation of Large Language Models) can be used to divide the prompt instructions input by the large model into two parts: "Instruction" and "Input". Among them, "Instruction" specifies the category of knowledge points that the model needs to judge, and "Input" is a sample of exercise text. This article combines "Instruction" and "Input" to form a prompt instruction (Prompt) to inspire the large model to generate appropriate answers. The final output result (Output) is the category ID of the corresponding knowledge point. The large model can be a ChatGLM model. The fine-tuning process can be formally expressed as:
[0051] Formula (1);
[0052] Formula (2);
[0053] S120, converting each component element in the exercise text sample into a high-dimensional vector, combining the context information of the exercise text sample to obtain text features, and obtaining question type features based on the feature elements in the exercise text sample.
[0054] Here, the exercise text samples can be segmented, that is, the exercise text samples are divided into a series of words or phrases. The pre-trained word embedding model (BERT) is used to convert the segmented words or phrases into high-dimensional vectors. These high-dimensional vectors can capture the semantic relationship between words, so that semantically similar words are also close in the vector space. After obtaining the vector representation of each sentence or paragraph, the high-dimensional vector can be further enriched with contextual information to obtain text features. In addition, by performing structural analysis on the exercise text, characteristic elements in the exercise text samples are obtained. Characteristic elements can indicate the question type to a certain extent. For example, in the knowledge points of "angle measurement" and "parallelogram and trapezoid", there are more drawing exercises; while in the knowledge points of "understanding of large numbers" and "hectares and square kilometers", there are more conversion exercises. These can be achieved through regular expressions, syntax analysis trees or special parsers.
[0055] In some embodiments, S120, each component element in the exercise text sample is converted into a high-dimensional vector, and combined with the context information of the exercise text sample to obtain text features, including:
[0056] Use the pre-trained model Bert as a feature extractor to convert each component element in the exercise text sample into a high-dimensional vector;
[0057] The high-dimensional vector is input into the bidirectional long short-term memory network, and the context information of the exercise text sample is integrated to obtain the feature vector of each component element;
[0058] The sum of the feature vectors of the constituent elements is determined as the text feature of the exercise text sample.
[0059] In this embodiment, the pre-trained model Bert is based on the Transformer architecture, and Bert can be regarded as a transformer with only the encoder part. Bert is composed of a group of N encoders connected in series, and its standard configuration Bert-base model is composed of 12 layers of encoders connected in series, each of which contains a multi-head attention layer and a feedforward network layer. In Bert, the text of the exercise text sample is first segmented by the WordPiece word meta-analyzer. The character tokens are then sent to the token embedding layer, the segment embedding layer, and the position embedding layer, and the embedded values are obtained and added respectively, and then the results are sent to the encoder of the first layer for calculation. After being processed layer by layer by the encoder, the last layer will output a set of feature vectors of all characters. ,in It represents the first Here, the feature vector set includes multiple sets of high-dimensional vectors.
[0060] On this basis, the feature vector set is input into the bidirectional long short-term memory network Bi-LSTM, so that the feature vectors of all characters can be integrated with the context information of the sample sentences of the exercise text. Bi-LSTM consists of two independent LSTM networks, and the exercise text can be input into the Bi-LSTM network in forward and reverse order respectively. characters, we can get hidden state output in two directions and .Will and Splice to get Updated feature vector of characters As shown in formula (3), formula (4) and formula (5).
[0061] Formula (3);
[0062] Formula (4);
[0063] Formula (5);
[0064] The text features of the exercise text sample can be obtained by averaging all character feature vectors, as shown in formula (6).
[0065] Formula (6);
[0066] In some embodiments, S120, obtaining question type features based on characteristic elements in the exercise text sample includes:
[0067] Perform word segmentation on the exercise text samples to obtain multiple feature elements;
[0068] Based on different characteristic elements and the association between characteristic elements, question type characteristics are obtained.
[0069] In this embodiment, the question type feature may include question type, such as multiple choice question, fill-in-the-blank question, solution question, geometry question, etc. The question type of the exercise text sample can be represented by a one-hot vector. The vector is sent to a fully connected layer to obtain the deep question type feature of the exercise, as shown in formula (7).
[0070] Formula (7);
[0071] in, One-hot vector representing the question type of the exercise text sample, and are the learnable weight matrix and bias vector parameters.
[0072] S130, a feature vector is obtained based on the fusion of text features and question type features, and a preset classifier is trained using the feature vector and weak labels to obtain a target classifier.
[0073] After the text features and question type features of the exercise text sample are generated, the feature vector of the exercise text sample can be concatenated. The feature vector is input into a fully connected neural network, and the prediction probability of each knowledge point can be determined by calculating the Softmax activation function, as shown in formula (8).
[0074] Formula (8);
[0075] in, Indicates exercise text sample Belong to knowledge point The probability of and are the learnable weight matrix and bias vector parameters.
[0076] In some embodiments, S130, a feature vector is obtained based on the fusion of text features and question type features, and a preset classifier is trained using the feature vector and weak labels to obtain a target classifier, including:
[0077] Input the feature vector and weak label into the preset classifier, and output the mapping relationship between the feature vector and the weak label;
[0078] Adjust the parameters of the preset classifier based on the loss function corresponding to the mapping relationship until the loss function reaches a preset threshold;
[0079] The preset classifier when the loss function reaches a preset threshold is determined as the target classifier.
[0080] In this embodiment, the feature vector of the data set is used as input, and the weak label is sent to the preset classifier as the expected output. The difference between the predicted value of the classifier and the weak label is quantified using the loss function, and the parameters of the preset classifier are adjusted by minimizing the loss function. Exemplarily, it can be usually implemented by the gradient descent method or its variant methods (such as stochastic gradient descent, Adam optimizer, etc.). During the training process, the preset classifier will continuously iterate and update its parameters to reduce the difference between the predicted value and the weak label until the loss function reaches a preset threshold and stops iterating.
[0081] S140, inputting the exercise text to be predicted into the target classifier to determine the predicted knowledge points corresponding to the exercise text to be predicted.
[0082] The feature vectors of the exercise text are diverse, and correspondingly, the types of knowledge points are also diverse. After the exercise text to be predicted is input into the target classifier, at least one type of predicted knowledge point that matches the exercise text to be predicted can be obtained.
[0083] In some embodiments, S140, inputting the exercise text to be predicted into the target classifier, determining the predicted knowledge points corresponding to the exercise text for each group of knowledge points to be predicted by the prediction probability prediction, includes:
[0084] Input the exercise text to be predicted into the target classifier to obtain at least one set of candidate knowledge points;
[0085] A group of candidate knowledge points with the highest prediction probability is determined as the predicted knowledge points corresponding to the exercise text to be predicted.
[0086] In this embodiment, the probabilities of the predicted knowledge points can be sorted, and only a group of candidate knowledge points with the largest predicted probabilities are displayed as predicted knowledge points. Of course, the candidate knowledge points can also be displayed in descending order of predicted probabilities, and there is no limitation on this.
[0087] In one example, a set of exercises for the first semester of fourth grade mathematics in primary school was created using the learning resources in the People's Education Press elementary school mathematics textbooks, as well as the supporting teaching materials such as the "Five Three Days Practice" and "Dian Zhongdian" exercise books. There are 8 course units in this semester, including "Understanding of Large Numbers", "Hectares and Square Kilometers", "Measurement of Angles", "Multiplication of Three-Digit Numbers by Two-Digit Numbers", "Parallelograms and Trapezoidal Shapes", "Division with Two-Digit Divisors", and "Mathematical Wide Angle - Optimization". Here, 1,732 math problems were collected as samples, of which 550 samples were used to fine-tune the ChatGLM2-6B large model. The remaining samples are divided into training set, validation set, and test set in a ratio of 70:15:15. It should be noted that the labels of the training set and validation set used by the feature fusion module are all weak labels annotated by the large model.
[0088] In the weak label generation stage, the learning rate can be set to 5e-5, the number of training rounds is set to 20, and the batch size is set to 4. In the feature fusion stage, the maximum length of the exercises is set to 100, the learning rate is set to 1e-5, the number of training rounds is set to 30, the batch size is set to 4, the number of Bi-LSTM network layers is set to 3 layers, the number of hidden units in each layer is set to 256, the question embedding layer dimension is set to 48, and the Dropout is set to 0.2. The evaluation indicators use four indicators, namely, accuracy, precision, recall, and F1-score, to evaluate the performance of the method. According to the evaluation results corresponding to the evaluation indicators of this idea, it can be determined that the knowledge point prediction method provided by the present invention not only meets the actual needs in terms of evaluation accuracy, but also has lower dependence on manual annotation and stronger applicability.
[0089] This embodiment trains the classifier by utilizing the weak labels, text features, and question type features of the exercise text, fully considering the feature types of the exercise text, thereby improving the training effect of the classifier, and further improving the accuracy and reliability of the predicted knowledge points.
[0090] In some embodiments, see Figure 2 , Figure 2 The schematic diagram of the structure of the knowledge point prediction device provided in the embodiment of the present application; the knowledge point prediction device 200 provided in the embodiment of the present application includes: a weak label generation module 210, a feature extraction module 220, a training module 230 and a determination module 240, wherein:
[0091] The weak label generation module 210 is configured to match the exercise text samples with predefined keywords to determine the weak labels corresponding to each exercise text sample; the keywords represent different knowledge categories;
[0092] The feature extraction module 220 is configured to convert each component element in the exercise text sample into a high-dimensional vector, obtain text features in combination with context information of the exercise text sample, and obtain question type features based on the feature elements in the exercise text sample;
[0093] The training module 230 is configured to obtain a feature vector based on the fusion of the text feature and the question type feature, and train a preset classifier using the feature vector and the weak label to obtain a target classifier;
[0094] The determination module 240 is configured to input the to-be-predicted exercise text into the target classifier and determine the predicted knowledge points corresponding to the to-be-predicted exercise text.
[0095] In some embodiments, the feature extraction module 220 is specifically configured as follows:
[0096] Use the pre-trained model Bert as a feature extractor to convert each component element in the exercise text sample into a high-dimensional vector;
[0097] The high-dimensional vector is input into the bidirectional long short-term memory network, and the context information of the exercise text sample is integrated to obtain the feature vector of each component element;
[0098] The sum of the feature vectors of the constituent elements is determined as the text feature of the exercise text sample.
[0099] In some embodiments, the feature extraction module 220 is specifically configured as follows:
[0100] Perform word segmentation on the exercise text samples to obtain multiple feature elements;
[0101] Based on different characteristic elements and the association between characteristic elements, question type characteristics are obtained.
[0102] In some embodiments, the training module 230 is specifically configured as follows:
[0103] Input the feature vector and weak label into the preset classifier, and output the mapping relationship between the feature vector and the weak label;
[0104] Adjust the parameters of the preset classifier based on the loss function corresponding to the mapping relationship until the loss function reaches a preset threshold;
[0105] The preset classifier when the loss function reaches a preset threshold is determined as the target classifier.
[0106] In some embodiments, the determination module 240 is specifically configured as follows:
[0107] Input the exercise text to be predicted into the target classifier to obtain at least one group of candidate knowledge points and the prediction probability of each group of knowledge points;
[0108] A group of candidate knowledge points with the highest prediction probability is determined as the predicted knowledge points corresponding to the exercise text to be predicted.
[0109] The knowledge point prediction device provided in the embodiment of the present application is based on the above-mentioned knowledge point prediction method, which will not be described again here to avoid repetition.
[0110] It should be noted that the knowledge point prediction device provided in the embodiment of the present application and the knowledge point prediction method provided in the embodiment of the present application are based on the same application concept, so the specific implementation of this embodiment can refer to the implementation of the aforementioned knowledge point prediction method, and the repeated parts will not be repeated.
[0111] In some embodiments, see Figure 3 , Figure 3The electronic device 300 provided in the embodiment of the present application includes a processor 310 and a memory 320; the memory 320 stores a computer program, wherein the computer program implements the above-mentioned knowledge point prediction method when executed by the processor.
[0112] Specifically, the processor 310 may include, for example, a general-purpose microprocessor, an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 310 may also include an onboard memory for cache purposes. The processor 310 may be a single processing unit or multiple processing units for executing different actions of the method flow according to the embodiment of the present application.
[0113] The memory 320 may be any medium capable of containing, storing, conveying, propagating or transmitting instructions. For example, the memory 320 may include, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared or semiconductor system, device, component or propagation medium. Specific examples of the memory 320 include: a magnetic storage device, such as a magnetic tape or a hard disk (HDD); an optical storage device, such as a compact disk (CD-ROM); a random access memory (RAM) or a flash memory; and / or a wired / wireless communication link.
[0114] The present application also provides a computer-readable medium on which a computer program is stored, and when the program is executed by a processor, the above-mentioned knowledge point prediction method is implemented. The computer-readable medium may be included in the device / apparatus / system described in the above-mentioned embodiment; or it may exist independently without being assembled into the device / apparatus / system. The above-mentioned computer-readable medium carries one or more programs, and when the above-mentioned one or more programs are executed, the method according to the embodiment of the present application is implemented.
[0115] According to an embodiment of the present application, a computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a computer-readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, optical cable, radio frequency signal, etc., or any suitable combination of the above.
[0116] Those skilled in the art will appreciate that the features described in the various embodiments and / or claims of the present application may be combined and / or combined in a variety of ways, even if such combinations or combinations are not explicitly described in the present application. In particular, without departing from the spirit and teachings of the present application, the features described in the various embodiments and / or claims of the present application may be combined and / or combined in a variety of ways. All of these combinations and / or combinations fall within the scope of the present application. Therefore, the scope of the present application should not be limited to the above-described embodiments, but should be determined not only by the appended claims, but also by the equivalents of the appended claims.
Claims
1. A knowledge point prediction method, It is characterized in that include: Using predefined keywords to match the exercise text samples, determining the weak labels corresponding to each of the exercise text samples; the keywords represent different knowledge categories; Convert each component element in the exercise text sample into a high-dimensional vector, combine the context information of the exercise text sample to obtain text features, and perform word segmentation on the exercise text sample to obtain multiple feature elements; obtain question type features based on the association relationship between different feature elements; A feature vector is obtained based on the fusion of the text feature and the question type feature, and the feature vector and the weak label are input into a preset classifier, and a mapping relationship between the feature vector and the weak label is obtained as an output; a parameter of the preset classifier is adjusted based on a loss function corresponding to the mapping relationship until the loss function reaches a preset threshold; and the preset classifier when the loss function reaches the preset threshold is determined as a target classifier; The exercise text to be predicted is input into the target classifier to determine the predicted knowledge points corresponding to the exercise text to be predicted.
2. The knowledge point prediction method according to claim 1, It is characterized in that The step of converting each component element in the exercise text sample into a high-dimensional vector and combining the context information of the exercise text sample to obtain text features includes: Using the pre-trained model Bert as a feature extractor, each component element in the exercise text sample is converted into a high-dimensional vector; Inputting the high-dimensional vector into a bidirectional long short-term memory network and fusing the context information of the exercise text sample to obtain a feature vector of each of the component elements; The sum of the feature vectors of the component elements is determined as the text feature of the exercise text sample.
3. The knowledge point prediction method according to claim 1, It is characterized in that The step of inputting the to-be-predicted exercise text into the target classifier and determining the predicted knowledge points corresponding to the to-be-predicted exercise text includes: Inputting the to-be-predicted exercise text into the target classifier to obtain at least one group of candidate knowledge points and the prediction probability of each group of the knowledge points; A group of candidate knowledge points with the largest prediction probability is determined as the predicted knowledge points corresponding to the exercise text to be predicted.
4. A knowledge point prediction device, It is characterized in that include: Weak label generation module, feature extraction module, training module and determination module, among which, The weak label generation module is configured to use predefined keywords to match the exercise text samples to determine the weak labels corresponding to each of the exercise text samples; the keywords represent different knowledge categories; The feature extraction module is configured to convert each component element in the exercise text sample into a high-dimensional vector, combine the context information of the exercise text sample to obtain text features, and perform word segmentation processing on the exercise text sample to obtain multiple feature elements; based on the correlation relationship between different feature elements, obtain question type features; The training module is configured to obtain a feature vector based on the fusion of the text feature and the question type feature, and input the feature vector and the weak label into a preset classifier, and output a mapping relationship between the feature vector and the weak label; adjust the parameters of the preset classifier based on the loss function corresponding to the mapping relationship until the loss function reaches a preset threshold; determine the preset classifier when the loss function reaches the preset threshold as the target classifier; The determination module is configured to input the to-be-predicted exercise text into the target classifier to determine the predicted knowledge points corresponding to the to-be-predicted exercise text.
5. The knowledge point prediction device according to claim 4, It is characterized in that The feature extraction module is specifically configured as follows: Using the pre-trained model Bert as a feature extractor, each component element in the exercise text sample is converted into a high-dimensional vector; Inputting the high-dimensional vector into a bidirectional long short-term memory network and fusing the context information of the exercise text sample to obtain a feature vector of each of the component elements; The sum of the feature vectors of the component elements is determined as the text feature of the exercise text sample.
6. An electronic device comprising a processor and a memory; the memory stores a computer program, in, When the computer program is executed by the processor, the computer program implements the knowledge point prediction method according to any one of claims 1 to 3.
7. A computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the steps of the method according to any one of claims 1 to 3 are implemented.
