Knowledge graph generation method, system, storage medium and electronic device

By extracting feature vectors from students' learning data and inputting neural network models to generate knowledge point weights and updating the knowledge graph, the problems of fixed learning paths and large manual workload in the existing technology are solved, and dynamic updates and work of personalized learning paths are realized. The amount reduction.

CN114297399BActive Publication Date: 2024-11-08IFLYTEK CO LTD
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202111403397.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-11-24
Publication Date
2024-11-08
Estimated Expiration
2041-11-24

AI Technical Summary

Technical Problem

The existing knowledge graph generation method leads to fixed learning paths, which is difficult to meet the personalized learning needs of different students, and requires a large number of manual labels to generate knowledge point labels, which has the problem of large manual workload.

Method used

By obtaining the learning data set of target users, extracting feature vectors and inputting the trained neural network model, generating knowledge point weights, updating the user's knowledge graph, meeting personalized learning needs, and reducing manual workload.

Benefits of technology

It realizes dynamic update of the knowledge graph based on user learning data, meets personalized learning needs, significantly reduces the workload of manual labeling, and improves the flexibility and personalization of the learning path.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114297399B_ABST
    Figure CN114297399B_ABST
Patent Text Reader

Abstract

The present invention provides a knowledge graph generation method, system, storage medium and electronic device, wherein the method includes: obtaining a learning data set of a target user within a preset time period, wherein the learning data set includes at least one learning data; performing feature extraction processing on the learning data in the learning data set to obtain a target feature vector, wherein the target feature vector is used to characterize the learning data learned by the target user; performing knowledge point analysis processing on the target feature vector using a trained neural network model to output the knowledge point weights corresponding to the learning data set; updating the knowledge graph of the target user according to the knowledge point weights, and outputting the updated target knowledge graph. The present invention can generate a knowledge graph through a neural network model, thereby meeting the personalized learning needs of the target user and significantly reducing the manual workload.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a knowledge graph generation method, system, storage medium and electronic device. Background Art

[0002] The existing learning path planning and resource recommendation methods adopt the form of traditional knowledge graphs, that is, teaching and research experts organize and summarize the knowledge points learned by students in the subject in advance, and establish a learning path from one knowledge point to another based on the examination syllabus and their own experience, thereby forming an overall knowledge graph. For example, when a student makes a mistake in the test question of the previous knowledge point, the system will recommend similar questions and related knowledge point explanations, as well as the test questions of the next knowledge point and the specific content of the knowledge point. However, the existing knowledge graph generation and use methods will cause the overall planned learning path to be relatively fixed, which is difficult to meet the different actual learning needs of different students, and it is necessary to generate knowledge point labels for all the test questions in the question bank through manual annotation, which involves a large amount of manual workload. Summary of the invention

[0003] The present invention provides a knowledge graph generation method, system, storage medium and electronic device to solve the problem in the prior art that the learning path brought by the traditional knowledge graph is relatively fixed and cannot adapt to the personalized learning needs of students.

[0004] In a first aspect, the present invention provides a method for generating a knowledge graph, the method comprising:

[0005] Acquire a learning data set of a target user within a preset time period, wherein the learning data set includes at least one item of learning data;

[0006] Performing feature extraction processing on the learning data in the learning data set to obtain a target feature vector, wherein the target feature vector is used to represent the learning data learned by the target user;

[0007] Using a trained neural network model to perform knowledge point analysis on the target feature vector to output the knowledge point weights corresponding to the learning data set;

[0008] The knowledge graph of the target user is updated according to the knowledge point weights, and the updated target knowledge graph is output.

[0009] In one embodiment of the present invention, feature extraction processing is performed on the learning data in the learning data set to obtain a target feature vector including:

[0010] Extracting text features from each item of learning data in the learning data set and inputting the text features into a pre-training model to obtain a first feature vector, wherein the first feature vector is used to represent information of each item of learning data;

[0011] Performing attention processing on the first feature vector to obtain a second feature vector representing the correlation between the learning data in the learning data set, wherein the second feature vector is used to characterize the relationship between each item of learning data;

[0012] Obtain a feature vector of the target knowledge graph and merge it with the second feature vector to obtain the target feature vector.

[0013] In one embodiment of the present invention, obtaining the target knowledge graph includes:

[0014] When the target knowledge graph is first acquired for the target user, the preset original knowledge graph is used as the target knowledge graph;

[0015] When the target knowledge graph is not obtained for the target user for the first time, the target knowledge graph that has been updated historically is used as the target knowledge graph.

[0016] In one embodiment of the present invention, extracting text features from each item of learning data in the learning data set and inputting the text features into a pre-training model to obtain a first feature vector includes:

[0017] The learning data contained in the learning data set is input into a trained pre-training model to extract text features from the unlabeled text data constituting the learning data and output the first feature vector with a preset dimension.

[0018] In one embodiment of the present invention, extracting text features from each item of learning data in the learning data set and inputting the text features into a pre-training model to obtain a first feature vector includes:

[0019] Extracting text features from a preset amount M of learning data and inputting the text features into the preset pre-training model;

[0020] Outputting the first feature vector after preprocessing by the pretrained model, where the dimension of the first feature vector is M*N;

[0021] Wherein, N represents the preset dimension, and M and N are positive integers.

[0022] In one embodiment of the present invention, performing attention processing on the first feature vector to obtain a second feature vector representing the correlation between the learning data in the learning data set includes:

[0023] The first eigenvector is input into an attention processing function, and the attention processing function performs normalization processing according to the first eigenvector and an adjustment factor to obtain the second eigenvector representing the correlation between the learning data in the learning data set.

[0024] In one embodiment of the present invention, the step of obtaining the feature vector of the target knowledge graph and merging it with the second feature vector to obtain the target feature vector includes:

[0025] Obtaining a feature vector of the target knowledge graph;

[0026] Merging the feature vector with the second feature vector to obtain the target feature vector, where the dimension of the target feature vector is M*(X+N);

[0027] Among them, M represents the amount of learning data, N represents the preset dimension of the pre-training model, X represents the dimension of the target knowledge graph, and X, M, and N are positive integers.

[0028] In one embodiment of the present invention, the target knowledge graph includes multiple knowledge points and a weight corresponding to each knowledge point.

[0029] In one embodiment of the present invention, the step of using a trained neural network model to perform knowledge point analysis on the target feature vector to output the knowledge point weights corresponding to the learning data set includes:

[0030] The target feature vector is input into a neural network model based on an encoder-decoder structure, wherein the neural network model is a neural network model trained using learning data of a preset user within a preset time period as training sample data.

[0031] In one embodiment of the present invention, the step of using a trained neural network model to perform knowledge point analysis on the target feature vector to output the knowledge point weights corresponding to the learning data set includes:

[0032] After the input target feature vector is processed by the encoder layer of the neural network model through self-attention and addition normalization, it is processed by the feedforward network and then added and normalized and output to the next encoder layer for processing until the last encoder layer outputs a vector;

[0033] The decoder layer of the neural network model performs self-attention processing and additive normalization processing on the vector output by the last encoder layer, performs encoding and decoding attention processing and additive normalization processing, passes through a feedforward network, performs additive normalization processing, and outputs to the next decoder layer for processing, until the last decoding layer outputs the knowledge point weight;

[0034] The neural network model includes an encoder and a decoder, the encoder includes at least one identical encoder layer, and the decoder includes at least one identical decoder layer.

[0035] In one embodiment of the present invention, the learning data set includes test question data and learning content data.

[0036] In a second aspect, the present invention further provides a knowledge graph generation system, which includes:

[0037] A data acquisition module, used to acquire a learning data set of a target user within a preset time period, wherein the learning data set includes at least one item of learning data;

[0038] A target feature vector generation module is used to perform feature extraction processing on the learning data in the learning data set to obtain a target feature vector, wherein the target feature vector is used to represent the learning data learned by the target user;

[0039] A weight generation module, used to perform knowledge point analysis on the target feature vector using a trained neural network model to output the knowledge point weights corresponding to the learning data set;

[0040] The knowledge graph updating module is used to update the knowledge graph of the target user according to the knowledge point weights and output the updated target knowledge graph.

[0041] In a third aspect, the present invention also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein when the processor executes the program, it implements any one of the knowledge graph generation methods described in the first aspect above.

[0042] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any one of the knowledge graph generation methods described in the first aspect above.

[0043] The technical solution provided by the present invention can obtain the learning data set of the target user, and generate a target feature vector based on the learning data set, and then input the target feature into the neural network model to generate knowledge point weights, and use the knowledge point weights to update the target knowledge graph. The present invention can generate a knowledge graph through a neural network model, thereby meeting the personalized learning needs of the target user and significantly reducing the manual workload. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other implementation methods can be obtained based on these drawings without paying creative work.

[0045] Figure 1 It is a schematic diagram of the process of the knowledge graph generation method provided by the present invention;

[0046] Figure 2 It is a schematic diagram of the steps of updating the knowledge graph provided by the present invention;

[0047] Figure 3 It is a schematic diagram of the attention processing step for extracting text features provided by the present invention;

[0048] Figure 4 is a schematic diagram of the principle of the attention processing steps provided by the present invention;

[0049] Figure 5 is a schematic diagram of the steps of generating a target feature vector provided by the present invention;

[0050] Figure 6 is a schematic diagram of the original knowledge graph manually formulated by experts provided by the present invention;

[0051] Figure 7 is a schematic diagram of a neural network model based on an encoder-decoder provided by the present invention;

[0052] Figure 8 is a schematic diagram of an updated target knowledge graph provided by the present invention;

[0053] Fig. 9 It is a schematic diagram of the structure of the knowledge graph generation system provided by the present invention;

[0054] Fig.10 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] The terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein.

[0057] Knowledge Graph is a graph-based knowledge representation and organization method. Its core component is the semantic network, in which nodes represent domain concepts (entities) and edges represent semantic relationships between concepts. Knowledge is present in the knowledge graph in the form of triples (head entity, predicate, tail entity). However, existing knowledge graphs are generally constructed manually or directly obtained by crawling manually edited knowledge graphs from online platforms, resulting in a relatively fixed learning path for the overall plan, and subsequent manual updates require a large amount of work.

[0058] The present invention provides a knowledge graph generation method, system, storage medium and electronic device, which can obtain a learning data set of a target user, generate a target feature vector based on the learning data set, and then input the target feature into a neural network model to generate a knowledge point weight, and use the knowledge point weight to update the target knowledge graph to meet the personalized learning needs of different users. The knowledge graph obtained according to the technical solution proposed in the present invention can be applied in a personalized learning system. For example, students can use smart terminal devices to log in to the personalized learning system, complete various test questions recommended by the system, and complete their own level ability test based on the test questions. Then the system evaluates the student's personal level ability according to the test results, and obtains the student's knowledge weaknesses based on the evaluation results, and then updates the knowledge graph according to the student's knowledge weaknesses to provide a personalized learning path for subsequent use.

[0059] Combine the following Figure 1-Figure 10 Describe the knowledge graph generation method, system, storage medium and electronic device of the present invention.

[0060] See also Figure 1 , Figure 2 , Figure 1 is a flow chart of the knowledge graph generation method provided by the present invention, Figure 2 Schematic diagram of the steps of updating the knowledge graph provided by the present invention. Figure 1-2 As shown, the knowledge graph generation method provided by the present invention includes:

[0061] Step 101: Acquire a learning data set of a target user within a preset time period, wherein the learning data set includes at least one item of learning data.

[0062] Exemplarily, the learning data set includes test data and learning content data. The test data may be test data of various subjects, such as mathematics test data, English test data, etc., and the learning content data may be data of knowledge points of various subjects, such as mathematics knowledge point data, English knowledge point data, etc.

[0063] For example, obtain the data of all practice test questions and learning content data of the current user in a certain period of time.

[0064] Step 102: Perform feature extraction processing on the learning data in the learning data set to obtain a target feature vector, where the target feature vector is used to represent the learning data that the target user has learned.

[0065] Step 103: Use the trained neural network model to perform knowledge point analysis on the target feature vector to output the knowledge point weights corresponding to the learning data set.

[0066] Step 104: update the target user's knowledge graph according to the knowledge point weights, and output the updated target knowledge graph.

[0067] The above steps 102 to 104 are described in detail below.

[0068] In the above step 102, the feature extraction process is performed on the learning data in the learning data set to obtain the target feature vector, including:

[0069] Step 1021: extract text features from each item of learning data in the learning data set and input the text features into a pre-training model to obtain a first feature vector, wherein the first feature vector is used to characterize information of each item of learning data.

[0070] Step 1022: Perform attention processing on the first feature vector to obtain a second feature vector, where the second feature vector is used to characterize the relationship between each learning data.

[0071] Step 1023, obtain the feature vector of the target knowledge graph and merge it with the second feature vector to obtain the target feature vector, and the target feature vector is used to represent the information of the knowledge point.

[0072] In the knowledge graph generation method proposed in the present invention, a pre-trained model is introduced to realize feature extraction of natural language to obtain the first feature vector, and a second feature vector representing the correlation between a preset number of test questions and / or learning contents is obtained through a self-attention processing mechanism, and then the second feature vector is merged with the feature vector of the target knowledge graph to obtain a target feature vector for inputting a trained neural network, which is then processed by the neural network and outputs knowledge point weights, and the target knowledge graph is updated using the knowledge point weights, so that the target user can subsequently perform learning that meets personalized needs according to the learning path defined by the updated target knowledge graph.

[0073] The above step 1021 is described in detail below.

[0074] Exemplarily, in the knowledge graph generation method proposed in the present invention, text features can be extracted according to a preset number M of test data, according to the list information of the questions, answers and answer analysis, and the extracted text features can be input into the pre-trained model to obtain a first feature vector.

[0075] For example, the data of all the practice questions and / or the learning content data of the current user in a certain time period are obtained and input into the pre-training model in chronological order. For example, assuming that a set of test papers includes 100 test questions, the text data and / or learning content data of 100 test questions (i.e., M=100) can be input at one time, and the knowledge points contained in each set of test papers are manually marked on the page.

[0076] It should be noted that the pre-trained model proposed in the present invention does not need to manually label the test questions, and can directly input the text data of the current user's historical test questions.

[0077] Exemplarily, the preset pre-training models proposed in the present invention include but are not limited to ELMo (Embeddings from Language Models, ELMo for short) model, GPT (Generative Pre-training Transformer, GPT for short) model, BERT (Bidirectional Encoder Representation from Transformers, BERT for short) model, etc.

[0078] Exemplarily, in one implementation of the present invention, the pre-training model uses a BERT model, which is a language representation model for pre-training. The BERT pre-training model has the following advantages:

[0079] First, MLM (Masked Language Model) can be used to pre-train bidirectional Transformers (which is a Seq2Seq model consisting of an encoder and a decoder) to generate deep bidirectional language representations.

[0080] Second, after pre-training, only an additional output layer needs to be added for fine-tuning to achieve better performance in a variety of downstream tasks. In this process, there is no need to make task-specific structural modifications to BERT.

[0081] It can be seen from the above that since the present invention adopts the BERT pre-training model, the effect of natural language processing can be greatly improved.

[0082] Exemplarily, in one implementation of the present invention, a first feature vector is output after preprocessing by the pretrained model, and the dimension of the first feature vector is M*N;

[0083] Wherein, M represents the amount of learning data, N represents the preset dimension of the pre-training model, and M and N are positive integers.

[0084] Exemplarily, in one implementation of the present invention, step 1021 includes: inputting the learning data contained in the learning data set into a pre-trained model to extract text features from the unlabeled text data constituting the learning data and outputting a first feature vector with a preset dimension.

[0085] For example, assuming that the preset dimension N of the BERT pre-training model is 1024 dimensions, and the number of input learning data M is 100 test questions, then the first feature vector output by the BERT pre-training model has a dimension of 100*1024.

[0086] For example, in the technical solution disclosed in the present invention, since the M vectors in the first feature vector are independent of each other, attention processing is required so that the M vectors (for example, test questions) can be associated with each other, such as Figure 3 As shown in the figure, start represents the start tag, context1~N represents text, Answer1~N represents the answer, Answer key1~N represents the answer analysis, Delim represents concatenation, and Extract represents the action, that is, the action of extracting BERT text features.

[0087] The above-mentioned step 1022 is described in detail below.

[0088] See also Figure 4 , Figure 4Schematic diagram of the principle of the attention processing step provided by the present invention. Exemplarily, the technical solution disclosed by the present invention uses a scaled dot-product attention method to perform similarity calculation. Figure 4 As shown, the scaled dot product attention mechanism includes performing dot product, scaling, masking (optional operation), and normalization operations on the first and second input vectors, and performing dot product of the result of the normalization operation with the third vector.

[0089] Exemplarily, in one implementation of the present invention, step 1022 includes:

[0090] The first feature vector is input into an attention processing function, and the attention processing function performs normalization processing according to the first feature vector and an adjustment factor to obtain the second feature vector representing the correlation between the learning data in the learning data set.

[0091] The attention processing function can be described as mapping a query and a set of key-value pairs to an output, where query, key, value, and output are all vectors. The output is calculated as the weighted sum of the values, where the weight assigned to each value is calculated by the compatibility function of the query and the corresponding key. A single-headed attention processing function is represented by the following formula:

[0092]

[0093] Among them, the input values ​​of Q, K, and V are all the first eigenvectors, d k represents the adjustment factor, and softmax represents the normalization function.

[0094] When the first eigenvector of the input is large, the softmax function may have a very small gradient and is difficult to learn effectively. Therefore, the embodiment of the present invention adds d k The adjustment factor is adjusted to prevent the dot product of Q and K from being too large.

[0095] For example, the softmax normalization function can be an extension of the logistic function for multi-class problems, that is,

[0096] For example, in one implementation of the present invention, the vectors of dimension [100*1024] formed by the text features of 100 test questions are subjected to mutual attention processing to obtain a second feature vector having correlation between the 100 test questions, and the dimension of the second feature vector is [100*1024].

[0097] The above step 1023 is described in detail below.

[0098] See also Figure 5 , Figure 5 Schematic diagram of the step of generating a target feature vector provided by the present invention. Exemplarily, the step 1023 includes step 10231, obtaining the target knowledge graph.

[0099] Exemplarily, the acquiring the target knowledge graph includes:

[0100] When the target knowledge graph is first acquired for the target user, a preset original knowledge graph (for example, an original knowledge graph manually formulated by an expert) is used as the target knowledge graph;

[0101] When the target knowledge graph is not obtained for the target user for the first time, the target knowledge graph that has been updated historically (for example, last updated) is used as the target knowledge graph.

[0102] For example, the default weight of each knowledge point in the original knowledge graph is 0 (e.g. Figure 6 shown).

[0103] It should be noted that the purpose of merging the second feature vector with the feature vector of the target knowledge graph to obtain the target feature vector is to merge the knowledge point features of the target knowledge graph with the features of the learning data. This is conducive to matching the knowledge point weights output by the neural network with the knowledge points of the knowledge graph.

[0104] For example, Figure 6 is a schematic diagram of the knowledge graph manually formulated by experts proposed in the present invention, such as Figure 6 As shown, exemplarily, the knowledge graph manually formulated by the expert includes first-level knowledge points, each of the first-level knowledge points contains second-level knowledge points, each of the second-level knowledge points contains third-level knowledge points, and a weight corresponding to each of the third-level knowledge points.

[0105] Exemplarily, the knowledge graph may be a knowledge graph of various disciplines. For example, the first-level knowledge point of the knowledge graph of mathematics is the four arithmetic operations, and the first-level knowledge point contains second-level knowledge points. For example, the four arithmetic operations include addition, subtraction, multiplication, division, and the order of mixed operations; each second-level knowledge point contains third-level knowledge points. For example, the "addition" knowledge point contains the "sum = addend + addend" knowledge point and the "addend = sum - another number" knowledge point.

[0106] It should be noted that the content and level of the first-level knowledge points, second-level knowledge points, and third-level knowledge points can be formulated according to the specific circumstances of each subject, and the present invention is not limited to Figure 6 The knowledge points of the mathematics subject "Four Arithmetic Operations" and an example of dividing the knowledge points into three levels are shown.

[0107] Step 10232: merge the feature vector with the second feature vector to obtain the target feature vector, where the dimension of the target feature vector is M*(X+N).

[0108] Among them, M represents the amount of learning data, X represents the dimension of the target knowledge graph, N represents the preset dimension of the pre-training model, and X, M, and N are positive integers.

[0109] Exemplarily, assuming that the dimension X of the target knowledge graph is 2018, the number of test questions M is 100, and N is 1024 dimensions, then after merging the feature vector of the target knowledge graph with the second feature vector, a target feature vector is obtained, and the dimension of the target feature vector is 100*(2018+1024), that is, 100*3042.

[0110] Exemplarily, the embodiment of the present invention may also convert the second feature vector into a high-dimensional feature, and then merge the feature vector with the second feature vector to obtain the target feature vector, and the target feature vector is used as the input of the neural network model.

[0111] The above step 103 is described in detail below.

[0112] Exemplarily, the aforementioned step 103 includes:

[0113] The target feature vector is analyzed and processed by a trained neural network model to output the knowledge point weights corresponding to the learning data set. The neural network model is a neural network model trained using the learning data of a preset user (e.g., a user whose performance has improved significantly) within a preset period of time as training sample data.

[0114] For example, the user's practice test data and learning content data sequence of the previous time period (which can be one year or one month) of the user whose scores have significantly improved within a preset period of time, as well as the feature vector of the knowledge graph updated by the user last time are used as the input layer, and the knowledge points of the user's practice test data of the next year (which can be one year or one month) of the user shown are used as the output results of the output layer, and massive data are used to train the model to fit the overall neural network.

[0115] Exemplarily, the trained neural network model includes an encoder and a decoder.

[0116] See also Figure 7 , Figure 7is a schematic diagram of a neural network model based on an encoder-decoder provided by the present invention. The neural network model including an encoder-decoder (Encoder-Decoder) is a general framework, under which different algorithms can be used to solve different tasks. Exemplarily, the encoder is composed of at least one identical encoder layer (encoder #1 and encoder #2 are shown in the figure), and the decoder is composed of at least one identical decoder layer (decoder #1 and decoder #2 are shown in the figure). The present invention does not limit the number of layers of the encoder layer and the decoder layer.

[0117] like Figure 7 As shown, exemplarily, each encoder layer (e.g., encoder #1) includes a self-attention layer, an addition & normalization layer, a feedforward network layer, and an addition & normalization layer, and each decoder layer (e.g., decoder #1) includes a self-attention layer, an addition & normalization layer, an encoding-decoding attention layer, an addition & normalization layer, a feedforward network layer, and an addition & normalization layer.

[0118] In the above step 103, the use of the trained neural network model to perform knowledge point analysis on the target feature vector to output the knowledge point weights corresponding to the learning data set includes:

[0119] The target feature vector input is processed by the encoder layer of the neural network model through self-attention and additive normalization, and then processed by the feedforward network and then additively normalized and output to the next encoder layer for processing, that is, the next encoder layer also performs the above operations until the last encoding layer outputs a vector.

[0120] The decoder layer of the neural network model performs self-attention processing and additive normalization processing on the vector output by the last encoder layer, and then performs encoding and decoding attention processing and additive normalization processing. After being processed by the feedforward network, it is further subjected to additive normalization processing and output to the next decoder layer for processing, that is, the next decoder layer also performs the above operations until the last one outputs the weight of the knowledge point.

[0121] The encoder is used to process the input (such as text, pictures, audio) and output it in the form of a vector. The decoder is used to process the input (vector) and output it in the form of text. The neural network based on the encoder-decoder structure has a sequence as input and a sequence as output. The encoder converts a variable-length signal sequence into a fixed-length vector expression, and the decoder converts the fixed-length vector into a variable-length target signal sequence.

[0122] Exemplarily, the present invention can select different encoders and decoders according to different tasks. For example, but not limited to, the neural network model proposed in the present invention can adopt RNN (Recurrent Neural Network) model structure, LSTM (Long Short-Term Memory) model structure, or GRU (Gate Recurrent Unit) model structure, or Transformers model structure. Therefore, the present invention does not limit the neural network model of the encoder and decoder structure.

[0123] It should be noted that when the Transformers neural network model structure is used, the input values ​​of the Q, K, and V parameters in the attention processing functions in the encoder and decoder are all the target feature vectors. If the RNN model structure is used, the self-attention processing layer may not be set.

[0124] The above step 104 is described in detail below.

[0125] In the above step 104, the knowledge graph of the target user is updated according to the knowledge point weights, and the updated target knowledge graph is output.

[0126] For example, Figure 8 is a schematic diagram of the updated target knowledge graph of the present invention, such as Figure 8 As shown, the updated target knowledge graph includes the third-level knowledge points and the knowledge point weights corresponding to each third-level knowledge point, wherein the updated knowledge point weights correspond to Figure 2 The black circle shown in the “Updated target knowledge graph” on the right.

[0127] That is to say, according to the weights of knowledge points output by the neural network (such as Figure 8 The middle column of the table) updates the knowledge graph of the target user (such as Figure 8 The right column of the table is the updated knowledge point weights), and the updated target knowledge graph is output. As can be seen from the above, the technical solution disclosed by the present invention plans a personalized learning path by updating the target knowledge graph according to the target user's own learning characteristics, that is, it can ultimately realize personalized learning path formulation and resource recommendation based on the detection and analysis of individual weak knowledge points, thereby playing a significant role in reducing the burden and increasing efficiency of students, and avoiding students from blindly conducting a large number of question training.

[0128] The present invention also proposes a knowledge graph generation system. The knowledge graph generation system described below and the knowledge graph generation method described above can refer to each other.

[0129] See also Fig. 9 , Fig. 9 Schematic diagram of the knowledge graph generation system provided by the present invention. Fig. 9 As shown, the knowledge graph generation system 900 proposed in the present invention includes a data acquisition module 901, a target feature vector generation module 902, a weight generation module 903 and a knowledge graph update module 904.

[0130] Among them, the data acquisition module 901 is used to obtain the learning data set of the target user within a preset time period, and the learning data set includes at least one learning data. The target feature vector generation module 902 is used to perform feature extraction processing on the learning data in the learning data set to obtain a target feature vector, and the target feature vector is used to characterize the learning data learned by the target user; the weight generation module 903 is used to use a trained neural network model to perform knowledge point analysis processing on the target feature vector to output the knowledge point weight corresponding to the learning data set; the knowledge graph update module 904 is used to update the knowledge graph of the target user according to the knowledge point weight, and output the updated target knowledge graph.

[0131] Exemplarily, the target feature vector generating module 902 is further used for:

[0132] Extracting text features from each item of learning data in the learning data set and inputting the text features into a pre-training model to obtain a first feature vector, wherein the first feature vector is used to represent information of each item of learning data;

[0133] Performing attention processing on the first feature vector to obtain a second feature vector representing the correlation between the learning data in the learning data set, wherein the second feature vector is used to characterize the relationship between each item of learning data;

[0134] Obtain a feature vector of the target knowledge graph and merge it with the second feature vector to obtain the target feature vector.

[0135] Exemplarily, the target feature vector generating module 902 is further used for:

[0136] When the target knowledge graph is first acquired for the target user, a preset original knowledge graph is used as the target knowledge graph; and

[0137] When the target knowledge graph is not obtained for the target user for the first time, the target knowledge graph that has been updated historically is used as the target knowledge graph.

[0138] Exemplarily, the target feature vector generating module 902 is further used for:

[0139] The learning data contained in the learning data set is input into a trained pre-training model to extract text features from the unlabeled text data constituting the learning data and output the first feature vector with a preset dimension.

[0140] Exemplarily, the target feature vector generating module 902 is further used for:

[0141] Extracting text features from a preset amount M of learning data and inputting the text features into the pre-trained model;

[0142] Outputting the first feature vector after preprocessing by the pretrained model, where the dimension of the first feature vector is M*N;

[0143] Wherein, N represents the preset dimension of the pre-trained model, and M and N are positive integers.

[0144] Exemplarily, the target feature vector generating module 902 is further used for:

[0145] The first eigenvector is input into an attention processing function, and the attention processing function performs normalization processing according to the first eigenvector and an adjustment factor to obtain the second eigenvector representing the correlation between the learning data in the learning data set.

[0146] Exemplarily, the target feature vector generating module 902 is further used for:

[0147] Obtaining a feature vector of the target knowledge graph;

[0148] Merging the feature vector with the second feature vector to obtain the target feature vector, where the dimension of the target feature vector is M*(X+N);

[0149] Among them, M represents the amount of learning data, X represents the dimension of the target knowledge graph, N represents the preset dimension of the pre-training model, and X, M, and N are positive integers.

[0150] Exemplarily, the target knowledge graph includes multiple knowledge points and a weight corresponding to each knowledge point.

[0151] Exemplarily, the weight generating module 903 is further used for:

[0152] The target feature vector is input into a neural network model based on an encoder-decoder structure, wherein the neural network model is a neural network model trained using the learning data of users whose performance has significantly improved within a preset period of time as training sample data.

[0153] Exemplarily, the weight generating module 903 is further used for:

[0154] After the input target feature vector is processed by the encoder layer of the neural network model through self-attention and addition normalization, it is processed by the feedforward network and then added and normalized and output to the next encoder layer for processing until the last encoder layer outputs a vector;

[0155] The decoder layer of the neural network model performs self-attention processing and additive normalization processing on the vector output by the last encoder layer, performs encoding and decoding attention processing and additive normalization processing, passes through a feedforward network, performs additive normalization processing, and outputs to the next decoder layer for processing, until the last decoding layer outputs the knowledge point weight;

[0156] The neural network model includes an encoder and a decoder, the encoder includes at least one identical encoder layer, and the decoder includes at least one identical decoder layer.

[0157] Exemplarily, the learning data set includes test question data and learning content data.

[0158] Other aspects of the knowledge graph generation system proposed in the present invention are the same or similar to the knowledge graph generation method described above and will not be repeated here.

[0159] On the other hand, the present invention also proposes an electronic device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein when the processor executes the program, it implements any one of the knowledge graph generation methods described above.

[0160] Fig.10 An example of a structural diagram of an electronic device is shown in FIG. Fig.10 As shown, the electronic device may include: a processor (Processor) 1010, a communication interface (Communications Interface) 1020, a memory (Memory) 1030 and a communication bus 1040, wherein the processor 1010, the communication interface 1020, and the memory 1030 communicate with each other through the communication bus 1040. The processor 1010 can call the logic instructions in the memory 1030 to execute the knowledge graph generation method

[0161] Exemplarily, the logic instructions in the above-mentioned memory 1030 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk.

[0162] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform any of the above-mentioned knowledge graph generation methods.

[0163] The embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or they may be 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. Those of ordinary skill in the art may understand and implement it without creative effort.

[0164] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0165] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A knowledge graph generation method, characterized in that: The method comprises: Acquire a learning data set of a target user within a preset time period, wherein the learning data set includes at least one item of learning data, and the learning data set includes test question data and learning content data; Performing feature extraction processing on the learning data in the learning data set to obtain a target feature vector, wherein the target feature vector is used to characterize the learning data learned by the target user, including: extracting text features from each learning data item in the learning data set, and inputting the text features into a pre-trained model to obtain a first feature vector, wherein the first feature vector is used to characterize the information of each learning data item; performing attention processing on the first feature vector to obtain a second feature vector, wherein the second feature vector is used to characterize the relationship between each learning data item; obtaining a feature vector of a target knowledge graph and merging it with the second feature vector to obtain the target feature vector; Using a trained neural network model to perform knowledge point analysis on the target feature vector to output the knowledge point weights corresponding to the learning data set; The target knowledge graph of the target user is updated according to the knowledge point weights, and the updated target knowledge graph is output.

2. The knowledge graph generation method according to claim 1, characterized in that: The acquiring of the target knowledge graph comprises: When the target knowledge graph is obtained for the target user for the first time, the preset original knowledge graph will be used as the target knowledge graph; when the target knowledge graph is not obtained for the target user for the first time, the historically updated target knowledge graph will be used as the target knowledge graph.

3. The knowledge graph generation method according to claim 2, characterized in that: The step of extracting text features from each item of learning data in the learning data set and inputting the text features into a pre-training model to obtain a first feature vector comprises: The learning data contained in the learning data set is input into a trained pre-training model to extract text features from the unlabeled text data constituting the learning data and output the first feature vector with a preset dimension.

4. The knowledge graph generation method according to claim 3, characterized in that: The step of extracting text features from each item of learning data in the learning data set and inputting the text features into a pre-training model to obtain a first feature vector comprises: Extracting text features from a preset amount M of learning data and inputting the text features into the pre-trained model; Outputting the first feature vector after preprocessing by the pretrained model, where the dimension of the first feature vector is M*N; Wherein, N represents the preset dimension of the pre-trained model, and M and N are positive integers.

5. The knowledge graph generation method according to claim 1, characterized in that: The performing attention processing on the first feature vector to obtain a second feature vector representing the correlation between the learning data in the learning data set comprises: The first eigenvector is input into an attention processing function, and the attention processing function performs normalization processing according to the first eigenvector and an adjustment factor to obtain the second eigenvector representing the correlation between the learning data in the learning data set.

6. The knowledge graph generation method according to claim 1, characterized in that: The acquiring the feature vector of the target knowledge graph and merging it with the second feature vector to obtain the target feature vector includes: Obtaining a feature vector of the target knowledge graph; Merging the feature vector with the second feature vector to obtain the target feature vector, where the dimension of the target feature vector is M*(X+N); Among them, M represents the amount of learning data, X represents the dimension of the target knowledge graph, N represents the preset dimension of the pre-training model, and X, M, and N are positive integers.

7. The knowledge graph generation method according to claim 1, characterized in that: The target knowledge graph includes multiple knowledge points and the weight corresponding to each knowledge point.

8. The knowledge graph generation method according to claim 1, characterized in that: The step of using the trained neural network model to perform knowledge point analysis on the target feature vector to output the knowledge point weights corresponding to the learning data set includes: The target feature vector is input into a neural network model based on an encoder-decoder structure, wherein the neural network model is a neural network model trained using learning data of a preset user within a preset time period as training sample data.

9. The knowledge graph generation method according to claim 1, characterized in that: The step of using the trained neural network model to perform knowledge point analysis on the target feature vector to output the knowledge point weights corresponding to the learning data set includes: After the input target feature vector is processed by the encoder layer of the neural network model through self-attention and addition normalization, it is processed by the feedforward network and then added and normalized and output to the next encoder layer for processing until the last encoder layer outputs a vector; The decoder layer of the neural network model performs self-attention processing and additive normalization processing on the vector output by the last encoder layer, and then performs encoding and decoding attention processing and additive normalization processing, and then performs additive normalization processing after processing through a feedforward network and then outputs it to the next decoder layer for processing, until the last decoding layer outputs the knowledge point weight after processing through a linear layer and a normalization layer; The neural network model includes an encoder and a decoder, the encoder includes at least one identical encoder layer, and the decoder includes at least one identical decoder layer.

10. A knowledge graph generation system, characterized in that: include: A data acquisition module, used to acquire a learning data set of a target user within a preset time period, wherein the learning data set includes at least one item of learning data, and the learning data set includes test question data and learning content data; A target feature vector generation module is used to perform feature extraction processing on the learning data in the learning data set to obtain a target feature vector, and the target feature vector is used to characterize the learning data learned by the target user, including: extracting text features from each learning data in the learning data set, and inputting the text features into a pre-training model to obtain a first feature vector, and the first feature vector is used to characterize the information of each learning data; performing attention processing on the first feature vector to obtain a second feature vector, and the second feature vector is used to characterize the relationship between each learning data; obtaining a feature vector of a target knowledge graph and merging it with the second feature vector to obtain the target feature vector; A weight generation module, used to perform knowledge point analysis on the target feature vector using a trained neural network model to output the knowledge point weights corresponding to the learning data set; The knowledge graph updating module is used to update the target knowledge graph of the target user according to the knowledge point weights and output the updated target knowledge graph.

11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: When the processor executes the program, it implements the knowledge graph generation method as described in any one of claims 1 to 9.

12. A processor-readable storage medium, characterized in that: The processor-readable storage medium stores a computer program, and the computer program is used to enable the processor to execute the knowledge graph generation method as described in any one of claims 1 to 9.

Citation Information

Patent Citations

  • Learning path pushing method and device, computer device and storage medium

    CN110263179A

  • Automatic text generation method and device and computer readable storage medium

    CN111651557A

  • Method and device for generating learning path based on knowledge graph, and storage medium

    CN112800236A

  • Method for detecting security of multi-dimensional data in networked cooperative system

    CN112822191A

  • Learning path planning method and device

    CN113610237A