Personalized Intelligent Question Answering Method and Device Based on Knowledge Tracing and Knowledge Graph

Abstract

The embodiment of the present invention discloses a personalized intelligent question-answering method and device based on knowledge tracking and knowledge graph, the method comprising: obtaining a student's question text, and obtaining a set of retrieval knowledge points of the question text; extracting a set of subject entities from the question text, retrieving data related to each subject entity from a preset knowledge graph and generating an answer text; obtaining relevant learning resources related to the question from a preset learning resource database according to the set of retrieval knowledge points, scoring and sorting the relevant learning resources and answer texts according to the personalized information of the students obtained from the student information database, and displaying the sorted relevant learning resources and answer texts to the students. Through the above-mentioned method, the embodiment of the present invention can accurately evaluate the students' mastery of knowledge points and learning preference characteristics, and display the answers and related learning resources generated by the knowledge graph in a personalized manner, so as to provide students with personalized and diversified intelligent question-answering services.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of intelligent education, and in particular, to a personalized intelligent question answering method and device based on knowledge tracing and knowledge graph. Background Art

[0002] Facing complex and diverse course contents, students often encounter various problems and confusions during the learning process. This not only poses high requirements for students' autonomous learning ability, but also when a large number of students have many questions that need to be answered by teachers, the huge tutoring workload may be unbearable for teachers. With the vigorous development and wide application of modern information technologies such as cloud computing, artificial intelligence, and big data mining in the education field, it provides strong support for solving the above problems. By using these technologies, an intelligent learning tutoring platform can be developed to provide real-time and personalized answers and guidance for specific problems of students.

[0003] The patent with the publication number CN108846104A discloses a question answering analysis and processing method and system based on an educational knowledge graph, which uses natural language processing algorithms to analyze the text of students' questions, extracts knowledge points, relationships, and the order information of knowledge points and relationships from the text, and combines the designed rules to retrieve the answers to the questions from the knowledge graph, and generates answer texts in the form of subject-predicate-object and feeds them back to students. The patent with the publication number CN110147436A discloses a hybrid automatic question answering method based on an educational knowledge graph and text, which respectively uses a text-based answer generation method and a knowledge graph-based answer generation method to generate answers to questions, and performs confidence scoring on each answer, and selects the answer to the question according to the confidence scoring result. Neither of the above two patents takes into account the students' mastery of knowledge points, and can only give the same answer to the same question.

[0004] It can be seen that the existing question answering methods for the education field usually only consider the matching situation between questions and answers, lack the utilization of students' mastery of knowledge points, and are difficult to provide accurate question answering services for students. In addition, the existing question answering methods for the education field usually answer students' doubts in the form of text, the form is too single, and does not consider the learning preference characteristics of students. Summary of the Invention

[0005] In view of the above problems, the embodiments of the present invention provide a personalized intelligent question answering method and device based on knowledge tracing and knowledge graph, which overcome the above problems or at least partially solve the above problems.

[0006] According to one aspect of the embodiments of the present invention, a personalized intelligent question answering method based on knowledge tracing and knowledge graph is provided. The method includes: obtaining the question text of a student, and obtaining the set of retrieved knowledge points of the question text; extracting a set of topic entities from the question text, retrieving data related to each of the topic entities from a preset knowledge graph, and generating an answer text; obtaining relevant learning resources related to the question from a preset learning resource database according to the set of retrieved knowledge points, respectively scoring and sorting the relevant learning resources and the answer text according to the personalized information of the student obtained from the student information database, and presenting the sorted relevant learning resources and the answer text to the student.

[0007] Optionally, before obtaining the question text of the student and obtaining the set of retrieved knowledge points of the question text, it includes: obtaining the cognitive state of the student according to the course selected by the student and the corresponding answering data, and saving it to the student information database; dividing the students into multiple categories based on the learning style theory using the K-means clustering algorithm according to the behavior data of each student, and saving it to the student information database; collecting a variety of learning resources according to the configuration document, processing the collected variety of learning resources to obtain the relevant text and embedding vectors of the learning resources, and storing the embedding vectors in the learning resource database; extracting triples including head entities, relationships, and tail entities according to the relevant text using the BERT-BiLSTM-CRF model, and storing them in the knowledge graph.

[0008] Optionally, obtaining the cognitive state of the student according to the course selected by the student and the corresponding answering data, and saving it to the student information database includes: pushing multiple exercise questions to the student according to the course selected by the student, and collecting the corresponding answering data of the student; obtaining the historical answering data of the student from the student information database, and splicing it with the answering data to obtain complete answering data; applying a deep knowledge tracing model to process the complete answering data, obtaining the student's mastery of each knowledge point, and obtaining the cognitive state of the student; storing the knowledge point mastery and the complete answering data in the student information database.

[0009] Optionally, processing the collected variety of learning resources to obtain the relevant text and embedding vectors of the learning resources includes: cleaning the collected variety of learning resources, and deleting invalid learning resources with problems; obtaining the relevant text of each cleaned learning resource; generating embedding vectors according to the relevant text using the BERT model; completing the knowledge point attributes of the learning resources according to the relevant text and the embedding vectors.

[0010] Optionally, extracting triples including a head entity, a relation, and a tail entity from the relevant text by applying the BERT-BiLSTM-CRF model and storing them in the knowledge graph includes: for any sentence in the relevant text, applying the BERT model to obtain word vectors of each word in the sentence; applying the BiLSTM model to process the word vectors of each word to obtain hidden state vectors of each word in the sentence; inputting the hidden state vectors of each word into a fully connected layer to calculate the probabilities of each word on all labels; predicting the optimal label sequence of the sentence according to the probabilities of each word on all labels by applying the CRF model; extracting the triples <head entity, relation, tail entity> of the knowledge graph according to the optimal label sequence and storing them in the knowledge graph.

[0011] Optionally, extracting a set of topic entities from the question text, retrieving data related to each of the topic entities from a preset knowledge graph, and generating an answer text includes: analyzing and processing the question text by applying the BERT-BiLSTM-CRF model to obtain a label sequence prediction result, and extracting each topic entity according to the label sequence prediction result to form a set of topic entities; retrieving each of the topic entities from the knowledge graph to obtain all relationships existing for each of the retrieved topic entities and entity nodes pointed to by the relationships; for any relationship, generating an answer text based on the answer template corresponding to the relationship in the configuration document according to the triple <head entity, relation, tail entity> obtained from the knowledge graph.

[0012] Optionally, scoring and sorting the relevant learning resources and the answer text according to the personalized information of the student obtained from the student information database, and presenting the sorted relevant learning resources and answer text to the student includes: obtaining the personalized information of the student from the student information database and calculating the first interest degree of the student in each resource in the relevant learning resources; calculating the mastery score of the student in each resource in the relevant learning resources according to the knowledge point mastery situation of the student obtained from the student information database; obtaining the first matching degree score of each resource in the relevant learning resources with the question text; performing a weighted sum of the first interest degree, the mastery score, and the first matching degree score to obtain the learning resource score of each resource in the relevant learning resources; obtaining the second interest degree of the student in the answer text and the second matching degree score of the answer text with the question text according to the personalized information of the student, and performing a weighted sum of the second interest degree and the second matching degree score to obtain the answer text score; sorting the relevant learning resources and the answer text from high to low according to the learning resource score and the answer text score and presenting them to the student.

[0013] Based on the same inventive concept, a personalized intelligent question-answering device based on knowledge tracing and knowledge graph is provided. A question analysis module is configured to obtain the question text of a student and obtain the set of retrieved knowledge points of the question text. An answer generation module is configured to extract the set of topic entities from the question text, retrieve the data related to each of the topic entities from a preset knowledge graph, and generate an answer text. A personalized question-answering module is configured to obtain relevant learning resources related to the question from a preset learning resource database according to the set of retrieved knowledge points, score and sort the relevant learning resources and the answer text respectively according to the personalized information of the student obtained from a student information database, and display the sorted relevant learning resources and the answer text to the student.

[0014] Based on the same inventive concept, an embodiment of the present invention further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the foregoing method is implemented.

[0015] Based on the same inventive concept, an embodiment of the present invention further provides a computer storage medium, in which at least one executable instruction is stored, and the executable instruction causes the processor to execute the foregoing method.

[0016] In the embodiment of the present invention, by obtaining the question text of a student and obtaining the set of retrieved knowledge points of the question text; extracting the set of topic entities from the question text, retrieving the data related to each of the topic entities from a preset knowledge graph, and generating an answer text; obtaining relevant learning resources related to the question from a preset learning resource database according to the set of retrieved knowledge points, scoring and sorting the relevant learning resources and the answer text respectively according to the personalized information of the student obtained from a student information database, and displaying the sorted relevant learning resources and the answer text to the student, it is possible to accurately evaluate the student's knowledge point mastery situation, and display the answer generated by the knowledge graph and relevant learning resources in a personalized manner according to the student's knowledge point mastery situation and learning preference characteristics, so as to provide personalized and diversified intelligent question-answering services for students.

[0017] The above description is only an overview of the technical solution of the embodiment of the present invention. In order to be able to understand the technical means of the embodiment of the present invention more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of the embodiment of the present invention more obvious and understandable, the following specifically enumerates the specific embodiments of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] By reading the following detailed description of the preferred embodiments, various other advantages and benefits will become apparent to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention. Moreover, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0019] Figure 1 shows a schematic structural diagram of a personalized intelligent Q&A device based on knowledge tracing and knowledge graph provided by an embodiment of the present invention;

[0020] Figure 2 shows a schematic flowchart of a personalized intelligent Q&A method based on knowledge tracing and knowledge graph provided by an embodiment of the present invention;

[0021] Figure 3 shows a schematic diagram of an electronic device in an embodiment of the present invention. Detailed Embodiments

[0022] The exemplary embodiments of the present invention will be described in more detail below with reference to the drawings. Although the exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present invention can be more thoroughly understood and the scope of the present invention can be fully communicated to those skilled in the art.

[0023] Figure 1 shows a schematic structural diagram of a personalized intelligent Q&A device based on knowledge tracing and knowledge graph provided by an embodiment of the present invention. As Figure 1 shown, the personalized intelligent Q&A device based on knowledge tracing and knowledge graph includes: a question analysis module, an answer generation module, and a personalized Q&A module. Among them,

[0024] The question analysis module is used to obtain the question text of the student and obtain the set of retrieved knowledge points of the question text;

[0025] The answer generation module is used to extract the set of topic entities from the question text, retrieve the data related to each of the topic entities from a preset knowledge graph, and generate an answer text;

[0026] The personalized Q&A module is used to obtain relevant learning resources related to the question from a preset learning resource database according to the set of retrieved knowledge points, score and sort the relevant learning resources and the answer text respectively according to the personalized information of the student obtained from the student information database, and display the sorted relevant learning resources and the answer text to the student.

[0027] In the embodiments of the present invention, before the problem analysis module obtains the problem text of the student and the retrieval knowledge point set of the problem text, it is necessary to construct a knowledge graph, a learning resource database, and a student information database. Thus, the personalized intelligent Q&A device based on knowledge tracing and knowledge graph in the embodiments of the present invention further includes: a cognitive state evaluation module and a learning style classification module for constructing a student information database, a learning resource collection module and a learning resource processing module for constructing a learning resource database, and a knowledge graph construction module for constructing a knowledge graph. Among them, the knowledge graph construction module is connected to the learning resource processing module.

[0028] In the embodiments of the present invention, the cognitive state evaluation module pushes practice questions to the student according to the courses selected by the student. After the student completes the practice questions, the knowledge tracing model is applied to analyze the student's answering data to obtain the student's cognitive state, that is, the mastery of different knowledge points. Finally, the cognitive state is saved to the student information database.

[0029] Specifically, first, according to the courses selected by the student, the practice questions are pushed to the student for answering. When the student completes the practice questions, the student's answering data is collected. Suppose the student answers n practice questions. In the answering data, the answering data of the i-th question is expressed as q i =(k i , a i ), where k i is the knowledge point set examined by the i-th question, and a i is the result of the student answering the i-th question. If the answer is correct, then a i = 1, otherwise a i = 0. Then, the student's past historical answering data is obtained from the student information database, and the newly collected answering data is spliced to obtain the complete answering data of the current student, so that the knowledge tracing model can accurately evaluate the mastery of knowledge points.

[0030] The complete answering data needs to be organized into the input form required by the knowledge tracing model. Taking the widely used Deep Knowledge Tracing (DKT) model as an example, assuming the total number of knowledge points is n, then for the answering data q i of each practice question, a vector x i with an initial value of 0 and a dimension of 2×n will be constructed first. If the question is answered correctly, that is, a i = 1, the k i -th dimension of the vector x i is modified to 1, otherwise the (n + k) i -th dimension of the vector x iThe dimension is modified to 1. The knowledge tracing model can evaluate the learning status of students based on the complete answering data of students. Taking the widely used DKT model as an example, DKT uses a Recurrent Neural Network (RNN) to evaluate the learning status of students. The formula is as follows:

[0031] y = RNN(X, W RNN )

[0032] K = MLP(y, W M )

[0033] where X is the complete sorted answering data; y is the output of the RNN; W RNN is the weight of the RNN; W M is the weight of the multi-layer perceptron MLP, and K = {η 1 , η 2 , …, η n} is the knowledge mastery situation of the student, that is, the cognitive state of the student. η i represents the mastery degree of the student for knowledge point i, and its value ranges from 0 to 1. Finally, the knowledge mastery situation K of the student and the complete answering data X are stored in the student information database.

[0034] The learning style classification module is based on the learning behavior data of students, introduces the learning style theory, and uses the K-means clustering algorithm to classify students into multiple categories. Students have different learning habits and preferences during the learning process. In the field of psychology, these differences in habits and preferences are called learning styles. In the embodiments of the present invention, the learning style theory is introduced, the learning style of students is identified according to the learning behavior of students, and students are classified according to the learning style, which is beneficial to improving the efficiency of personalized retrieval of learning resources. In the embodiments of the present invention, according to a preset first time period (for example, every 7 days or 10 days), the timer starts the learning style classification module to identify and update the learning style category to which the student belongs. The learning style classification module is used to model eight characteristics of students: dynamic, active, visual, verbal, sensory, intuitive, sequential, and holistic based on the Feld-Silverman model, where the learning behavior data includes the number of questions asked, the number of times of browsing learning resources, the learning duration, etc. Then, the K-means clustering algorithm is used to classify students into C categories. Among them, each student is represented by an eight-dimensional vector, and the value of the i-th dimension represents the i-th characteristic value of the student. Finally, the silhouette coefficient (SC) and the variance ratio criterion (Calinski-Harabasz, CH) index are used to evaluate the clustering effect, the C value with the best clustering effect is selected, and the classification result is saved to the student information database.

[0035] During the operation of the system, the timer will automatically trigger the learning resource collection module. The learning resource collection module reads the links and keywords in the configuration document, collects various learning resources such as course questions, course videos, course PPTs, and course documents from the Internet, and transfers these learning resources to the learning resource processing module. Specifically, in the embodiment of the present invention, according to a preset second time period (for example, every 24 hours or 48 hours), the timer starts the learning resource collection module, reads the data source and collection keywords from the configuration document, and collects various learning resources such as course questions, course videos, course PPTs, and course documents from the Internet. For different types of resources, the collected content is different. For example:

[0036] (1) Question collection: The module collects course question resources from various question banks on the Internet according to the content of the configuration document. The types of questions include multiple-choice questions, fill-in-the-blank questions, and short-answer questions. The collected content includes the content of the questions, answers, analysis, difficulty level, knowledge points, and corresponding courses;

[0037] (2) Course video collection: The module collects course video resources from various online learning platforms or other teaching resource libraries according to the content of the configuration document. The collected content includes the play link, title, description, preview picture, knowledge points, and corresponding courses of the course video;

[0038] (3) Course PPT collection: The module collects course PPT resources from online learning platforms or other teaching resource libraries according to the content of the configuration document. The collected content includes the title, browsing link, description, knowledge points, and corresponding courses of the course PPT;

[0039] (4) Course document collection: The module collects course document resources from the teaching resource library according to the content of the configuration document. The collected content includes the title, content, access link, knowledge points, and corresponding courses of the course document.

[0040] During the process of learning resource collection, the relevant rules of the collected website are followed to respect the intellectual property rights of others. These resources will then be transferred to the learning resource processing module for subsequent processing.

[0041] The learning resource processing module receives the learning resources transferred from the learning resource collection module. First, data cleaning will be performed, that is, invalid learning resources will be deleted. Then, the relevant text of the learning resources will be embedded into embedding vectors, and the knowledge point attributes of the learning resources will be completed. Finally, the learning resources and the embedding vectors will be stored in the learning resource database, and the relevant text will be transferred to the knowledge graph construction module for subsequent processing. In the embodiment of the present invention, there will be some problems in the data collected by the learning resource collection module, such as course videos without titles, course documents with empty content, etc. The learning resource processing module first needs to delete these invalid learning resources.

[0042] Then, the cleaned learning resources are embedded in the form of embedding vectors so that the resources can be retrieved accurately and quickly subsequently, and it can also assist in completing the matching between learning resources and knowledge points. Here, the relevant text of the learning resources is selected to be embedded as embedding vectors. Therefore, for each learning resource, its relevant text needs to be prepared first. For test question resources, their content is directly selected as the relevant text; for video resources, their title plus description is selected as the relevant text; for PPT resources and document resources, their title plus text content is selected as the relevant text.

[0043] Neural networks can learn complex non - linear relationships in text and capture rich information between relevant semantics and context. The learning resource processing module in the embodiments of the present invention applies the BERT model to embed the relevant text of learning resources. Suppose the relevant text of learning resource r is T r ={t r,1 , t r,2 , t r,3 , …, t r,n}, where t r,i is the i - th sentence in the relevant text of learning resource r. After operations such as word embedding, position embedding, and multi - layer Transformer encoding using the BERT model, the average of the word vector sequences output by the BERT model is taken, and it is used as the embedding vector x r,i of sentence t r,i .

[0044] Among the collected learning resources, the knowledge point attributes of some learning resources may be empty. In order to retrieve learning resources through knowledge points subsequently, it is necessary to complete the knowledge point attributes of learning resources. The embodiments of the present invention use the relevant text and embedding vectors of learning resources to assist in completing the supplementation of knowledge point attributes. First, relevant data of knowledge points are read from the configuration document, including relevant vocabulary of knowledge points (such as the name of knowledge points, synonyms of the name, etc.) and corresponding vectors of knowledge points. Then, for the relevant text, first use the jieba word - segmentation tool to segment the relevant text, compare the words obtained by word - segmentation with the relevant vocabulary of knowledge points, and add the knowledge points with overlapping vocabulary to the knowledge point attributes of this resource. Finally, for the embedding vectors, a multi - layer perceptron is used to calculate the matching degree between the embedding vectors and the corresponding vectors of knowledge points, and the knowledge points with a matching degree exceeding the threshold are added to the knowledge point attributes of this resource.

[0045] Finally, for learning resource r, its corresponding embedding vector set X r ={x r,1 , x r,2 , …, x r,n} It is stored in the learning resource database. And the relevant text of the learning resource is passed to the knowledge graph construction module to construct the knowledge graph.

[0046] The knowledge graph construction module receives the relevant text of the learning resource passed from the learning resource processing module. For a certain sentence t = {w 1 , w 2 , w 3 , …, w n}, where w i represents the i-th word of the sentence t, and n represents the number of words. Here, the BERT-BiLSTM-CRF model is applied to obtain the label sequence of the sentence t. First, the sentence t is input into the BERT model to get E = {e 1 , e 2 , e 3 , …, e n}, where e i is the word vector obtained by using the BERT model to embed w i . Long Short-Term Memory (LSTM) is a special recurrent neural network, and the bidirectional long short-term memory network (BiLSTM) consists of a forward LSTM and a backward LSTM. To further utilize the context information of the text, the knowledge graph construction module uses BiLSTM to process the output E of the BERT model, and the formula is as follows:

[0047]

[0048] where, and are the forward LSTM and the backward LSTM respectively, and h i is the hidden state vector of BiLSTM, which is composed of the forward hidden state vector and the backward hidden state vector concatenated.

[0049] Then, h i is input into the fully connected layer to calculate the probability p i of the word w i on all labels. p i is a vector, and its dimension size is the number of label types. This label contains three types of information: the first type is the position information of the entity word, that is, the position of the word in the entity vocabulary; the second type is the entity relationship category information, that is, the relationship category between this entity and another entity in the sentence; the third type is the position of this entity in the triple, that is, whether this entity is the head entity or the tail entity.

[0050] Since there is a dependency relationship between labels, assuming that the label corresponding to the word w i is b i, and label b i can only be followed by label b i+1 or label b i+2 , then for word w i+1 the label can only be b i+1 or b i+2 . Therefore, simply relying on the probability output by the fully connected layer cannot accurately determine the label to which the word belongs, while the Conditional Random Field (CRF) model can take into account the constraint relationship between adjacent labels, so as to obtain the globally optimal label sequence. Suppose the label sequence corresponding to sentence t is Y = {y 1 , y 2 , y 3 , …, y n}, then the score calculation formula for this sequence is as follows:

[0051]

[0052] Among them, is the probability that word w i is predicted to be label y i , is the probability that label y i-1 transfers to label y i . Then the Viterbi algorithm is used to search for the label sequence with the highest score to obtain the globally optimal label sequence.

[0053] Finally, according to the optimal label sequence predicted for sentence t, extract the triples <head entity, relation, tail entity> of the knowledge graph from the sentence text and store them in the knowledge graph.

[0054] So far, the construction of the knowledge graph, the learning resource database, and the student information database is completed. In the embodiments of the present invention, first, learning resources for a specific course are collected from the Internet, such as test questions, videos, courseware, etc., and the learning resources are analyzed to identify the knowledge points involved in the learning resources, and a learning resource database is constructed. Secondly, the BERT-BiLSTM-CRF model is applied for entity and relation recognition, and triples are extracted from the collected learning resources to construct a knowledge graph for a specific course. Then, the answering data of students is collected, and the knowledge tracing model is applied to diagnose the students' mastery of the knowledge points of a specific course; the learning style theory is introduced, the learning behavior data is used to model the learning style characteristics of students, and the K-means clustering algorithm is used to divide students into multiple categories to model the learning preference characteristics of students, so as to construct a student information database.

[0055] In the embodiment of the present invention, the problem analysis module is used to receive the questions asked by students, identify the knowledge points involved in the questions, and these knowledge points are called retrieval knowledge points. The question text and the retrieval knowledge points are passed to the answer generation module based on the knowledge graph for subsequent processing. There are three ways for students to input questions into the system. The first is to directly input the question text in the input box on the interaction page. The second is to click the recording button on the interaction page and input through voice. The third is to click the upload picture button on the interaction page and upload the question picture. When the student selects voice input, the device will record the student's voice and call a third-party speech-to-text service to convert the voice into text. When the student selects to upload a picture, the device receives the uploaded picture and calls an Optical Character Recognition (OCR) tool to extract the text content in the picture.

[0056] The problem analysis module inputs the text content of the question into the BERT model to obtain the embedding vector of the question text, and then inputs the embedding vector into a multi-layer perceptron to predict the knowledge points to which the question belongs, and adds the knowledge points to the retrieval knowledge point set. Then, the relevant vocabulary of the knowledge points is read from the configuration file, and the jieba word segmentation tool is used to segment the text content of the question, and the obtained vocabulary is compared with the relevant vocabulary of the knowledge points, and the knowledge points with overlapping vocabulary are added to the retrieval knowledge point set. Finally, the question text and the retrieval knowledge point set are passed to the answer generation module.

[0057] The answer generation module receives the question text W and the retrieval knowledge point set passed from the problem analysis module, and uses the BERT-BiLSTM-CRF model applied in the knowledge graph construction module to analyze and process the question text W to obtain the prediction result of the label sequence. According to the prediction result of the label sequence output by the model, the set of topic entities is extracted from the question text. The topic entity refers to the entity involved in the question, which may be one or more. For example, the topic entities of the question "What are the works of Li Bai and Du Fu?" are "Li Bai" and "Du Fu".

[0058] Then, according to the set of extracted topic entities, for a certain topic entity, retrieve its corresponding entity node from the knowledge graph, and obtain all the relationships existing in this node and the entity nodes pointed to by the relationships. For each relationship, read the corresponding answer template from the configuration file, and generate the answer text according to the triple <head entity, relationship, tail entity> obtained from the knowledge graph. For example, for a certain relationship "works" of the entity "Li Bai", the corresponding triples are <Li Bai, works, Thoughts on a Quiet Night>, <Li Bai, works, Viewing the Lu Mountain Waterfall>, <Li Bai, works, Song of the Moon over the Mountain Pass>, <Li Bai, works, Spending the Night in a Mountain Temple>, etc. Then, according to the template "{{author}}'s works include {{work list}}" corresponding to the relationship "works", generate the answer text: "Li Bai's works include Thoughts on a Quiet Night, Viewing the Lu Mountain Waterfall, Song of the Moon over the Mountain Pass, Spending the Night in a Mountain Temple, etc." Finally, pass the question text W, the set of answer texts, and the set of retrieved knowledge points to the personalized Q&A module for subsequent processing.

[0059] The personalized Q&A module receives the question text, retrieved knowledge points, and answer texts passed from the answer generation module. According to the retrieved knowledge points, retrieve relevant learning resources related to the question from the learning resource database, and score the relevant learning resources according to three factors: interest level, knowledge point mastery, and matching degree with the question. Score the answer texts according to two factors: interest level and matching degree with the question. Finally, sort the relevant learning resources and answer texts from high to low according to the scores and display them to the students.

[0060] Specifically, the personalized Q&A module receives the question text W, the set of answer texts, and the set of retrieved knowledge points passed from the answer generation module. Screen the learning resources from the learning resource database, and call the screened learning resources relevant learning resources. The screening condition is that the intersection of the knowledge point attributes of the learning resources and the set of retrieved knowledge points is not empty.

[0061] Among numerous learning resources, the more prominent the resource is ranked, the more quickly the student can notice it. To improve the effect of answering doubts, it is necessary to score the relevant learning resources according to three factors: the student's first interest level in the resource, the student's mastery of the knowledge points, and the matching degree of the resource with the question, so as to measure the matching degree between the learning resource and the student. The personalized Q&A module obtains the category c to which the student s belongs from the student information database i , then in the category c i , for any relevant learning resource r n and the learning resource r m , the similarity Sim(r n , r m |c i ) is defined as:

[0062]

[0063] Among them, S(r n |c i ) is the number of students in category c i who have viewed any relevant learning resource r n , r m is a learning resource in the set of learning resources viewed by student s. The numerator is the number of students in category c i who have viewed both any relevant learning resource r n and learning resource r m . Then, for student s, the first interest degree Int1(s, r n |c n ) is as follows: i ) is:

[0064]

[0065] Among them, L(s) is the set of learning resources viewed by student s.

[0066] The personalized Q&A module obtains the knowledge point mastery situation K s ={η s1 , η s2 , …, η sn} of the current student s from the student information database, where η si represents the mastery degree of student s for knowledge point i, with a value between 0 and 1. When students encounter problems that they cannot solve during the learning process, it may be because these problems involve relatively weak knowledge points for them. Therefore, when sorting learning resources, if a student has a worse mastery of the knowledge points involved in a certain resource, then this learning resource will be ranked higher in the sorting. Therefore, the formula for scoring the student's mastery of knowledge points is defined as:

[0067]

[0068] Among them, Kms(s, r n ) is the scoring of the mastery of the knowledge points involved in the relevant learning resource r n , represents the knowledge points involved in the relevant learning resource r n , that is, the knowledge point attribute of the relevant learning resource r n ; H W is the retrieved knowledge point; η sh represents the mastery situation of the current student s for knowledge point h.

[0069] In the embodiments of the present invention, the matching degree between the vectors in the set of embedding vectors of learning resources and the embedding vector of the question text is used as a measure of the matching situation between the learning resources and the current question. For any relevant learning resource r n The calculation formula for the first matching degree score with the question text is defined as:

[0070]

[0071] where is the set of embedding vectors of any relevant learning resource r n , where represents the vector corresponding to the j-th sentence of the relevant learning resource r n ; q is the embedding vector of the question text.

[0072] The weighted sum of the first interest degree, the mastery situation score, and the first matching degree score is calculated to obtain the learning resource score of each resource in the relevant learning resources. The calculation formula for the learning resource score is as follows:

[0073] score(s,r n ,q|c i )=l 1 ×Int1(s,r n |c i )+l 2 ×Kms(s,r n )+l 3 ×Rqs(q,r n )

[0074] where l 1 , l 2 , l 3 are the weight values of the first interest degree, the knowledge point mastery situation, and the matching situation between the question and the resource, and their magnitudes are determined by the importance of these three factors. According to the above formula, the score of each resource in the relevant learning resources is calculated, and the score is between 0 and 100.

[0075] When answering students' questions, some students prefer to browse learning resources, while some students prefer to browse the answers generated by the knowledge graph. Therefore, in the embodiments of the present invention, all answer texts will be combined with relevant learning resources and presented to students. The display order is determined by the scores of the answer texts and the learning resources. Therefore, it is necessary to incorporate the answer texts into a similar scoring system to measure the matching degree between the answer texts and the students. Referring to the scoring factors of relevant learning resources, the scoring factors for answer texts are: the second interest degree of students in the answer texts generated based on the knowledge graph and the matching situation between the answer texts and the question texts.

[0076] For the category c iFor student s, the second interest level in the answer text generated based on the knowledge graph is defined as:

[0077]

[0078] In the formula, N(c i ) represents the number of students in category c i , and P(c i ) represents the number of times students in category c i browsed the content generated based on the knowledge graph during past Q&A sessions.

[0079] For a certain answer text a, the calculation formula for its answer text score is as follows:

[0080] score(s,a,q|c i ) = l 4 × Int2(s|c i ) + l 5 × MLP(q,BERT(a))

[0081] In the above formula, l 4 , l 5 are the weight values of the second interest level and the second matching degree score, and their magnitudes are determined by the importance of these two factors; q is the embedding vector of the question text; MLP(q,BERT(a)) is the second matching degree score between the answer text a and the question text; BERT(a) is the embedding vector obtained by applying the BERT model to embed the answer text a. According to the above formula, the score of each answer text in the answer text set is calculated, and the score is between 0 and 100.

[0082] Finally, the personalized Q&A module combines the relevant learning resource set and the answer text set into one set, sorts them from high to low according to their scores, and displays the combined set in the form of a list on the interaction page with the student. When the student clicks on a certain learning resource, it jumps to the page corresponding to the resource; when the student clicks on a certain answer text, the full content of the answer text is displayed.

[0083] The personalized intelligent Q&A device based on knowledge tracing and knowledge graph according to the embodiments of the present invention recommends practice questions for students according to the courses selected by the students, collects the students' answering data, analyzes the answering data by applying a knowledge tracing model, and uses a recurrent neural network to identify potential features in the students' test score data, so as to accurately evaluate the students' mastery of knowledge points, and thus provide more matching learning resources for the students; compared with the existing Q&A systems in the education field, when conducting Q&A tutoring in the embodiments of the present invention, first analyze the students' question texts, generate answer texts by using a knowledge graph, and then, according to the students' knowledge point mastery and learning preference characteristics, personalizedly display the answers generated by the knowledge graph and related learning resources, and can provide personalized and diversified intelligent Q&A services for the students' weak knowledge points, so as to improve their learning effects.

[0084] For the convenience of description, the above device is described by dividing it into various modules according to functions. Of course, when implementing the embodiments of the present invention, the functions of each module can be implemented in the same or multiple software and / or hardware.

[0085] Figure 2 The flowchart of the personalized intelligent Q&A method based on knowledge tracing and knowledge graph provided by the embodiments of the present invention is shown. As Figure 2 shown, the personalized intelligent Q&A method based on knowledge tracing and knowledge graph includes:

[0086] Step S11: Obtain the students' question texts and obtain the set of retrieved knowledge points of the question texts.

[0087] In the embodiments of the present invention, before step S11, it is necessary to complete the construction of the student information database, the learning resource database, and the knowledge graph. First, obtain the students' cognitive states according to the courses selected by the students and the corresponding answering data, and save them in the student information database. Optionally, push multiple practice questions to the students according to the courses selected by the students, and collect the corresponding answering data of the students; obtain the students' historical answering data from the student information database, and splice it with the answering data to obtain complete answering data; apply a deep knowledge tracing model to process the complete answering data, obtain the students' mastery of each knowledge point, and obtain the students' cognitive states; store the knowledge point mastery and the complete answering data in the student information database. Also, divide the students into multiple categories based on the learning style theory by applying the K-means clustering algorithm according to the behavior data of each student, and save them in the student information database. Thus, the construction of the student information database is completed. For a more detailed process, refer to the device part above, and it will not be elaborated here.

[0088] Then, when constructing the learning resource database, various learning resources are collected according to the configuration document, and the collected various learning resources are processed to obtain the relevant text and embedding vectors of the learning resources, and the embedding vectors are stored in the learning resource database. When obtaining the relevant text and embedding vectors of the learning resources, optionally, the collected various learning resources are cleaned, and the invalid learning resources with problems are deleted; the relevant text of each cleaned learning resource is obtained; the BERT model is applied to generate embedding vectors according to the relevant text; the knowledge point attributes of the learning resources are complemented according to the relevant text and the embedding vectors. In this way, the construction of the learning resource database is completed. For a more detailed process, please refer to the previous device part and will not be elaborated here.

[0089] Finally, when constructing the knowledge graph, the BERT-BiLSTM-CRF model is applied according to the relevant text to extract triples including head entities, relations, and tail entities, and store them in the knowledge graph. Optionally, for any sentence in the relevant text, the BERT model is applied to obtain the word vectors of each word in the sentence; the BiLSTM model is applied to process the word vectors of each word to obtain the hidden state vectors of each word in the sentence; the hidden state vectors of each word are input into the fully connected layer to calculate the probabilities of each word on all labels; the CRF model is applied according to the probabilities of each word on all labels to predict the optimal label sequence of the sentence; the triples <head entity, relation, tail entity> of the knowledge graph are extracted according to the optimal label sequence and stored in the knowledge graph. In this way, the construction of the knowledge graph is completed. For a more detailed process, please refer to the previous device part and will not be elaborated here.

[0090] In step S11, the question asked by the student is received, and the question text of the student is obtained. There are three methods to input the question into the system. The first is to directly input the question text in the input box on the interaction page. The second is to click the recording button on the interaction page and input through voice. The third is to click the upload picture button on the interaction page and upload the question picture. When the student selects voice input, the student's voice is directly recorded, and a third-party voice-to-text service is called to convert the voice into text to obtain the question text; when the student selects to upload a picture, the uploaded picture is received, and an OCR tool is called to extract the text content in the picture to obtain the question text.

[0091] After obtaining the question text, the question text is input into the BERT model to obtain the embedding vector of the question text, and then the embedding vector is input into the multi-layer perceptron to predict the knowledge point to which the question belongs, and the knowledge point is added to the retrieved knowledge point set. For a more detailed process, please refer to the previous device part and will not be elaborated here.

[0092] Step S12: Extract a set of topic entities from the problem text, retrieve data related to each of the topic entities from a preset knowledge graph, and generate an answer text.

[0093] In an embodiment of the present invention, optionally, the BERT-BiLSTM-CRF model is applied to analyze and process the problem text to obtain a predicted result of a tag sequence, and each topic entity is extracted according to the predicted result of the tag sequence to form a set of topic entities; each of the topic entities is retrieved from the knowledge graph, and all relationships existing for each of the retrieved topic entities and entity nodes pointed to by the relationships are obtained; for any relationship, an answer text is generated based on an answer template corresponding to the relationship in the configuration document according to the triple <head entity, relationship, tail entity> obtained from the knowledge graph.

[0094] Step S13: Obtain relevant learning resources related to the problem from a preset learning resource database according to the retrieved knowledge point set, score and sort the relevant learning resources and the answer text respectively according to the personalized information of the student obtained from the student information database, and display the sorted relevant learning resources and the answer text to the student.

[0095] In an embodiment of the present invention, the problem text W, the set of answer texts, and the retrieved knowledge point set passed from the answer generation module are received. Learning resources are screened from the learning resource database, and the screened learning resources are called relevant learning resources. The screening condition is that the intersection of the knowledge point attributes of the learning resources and the retrieved knowledge point set is not empty.

[0096] Among numerous learning resources, the higher the ranking of a resource, the more quickly a student can notice it. When answering a student's question, some students prefer to browse learning resources, while some students prefer to browse the answers generated by the knowledge graph. Therefore, in an embodiment of the present invention, all answer texts and relevant learning resources are combined and displayed to the student. The display order is determined by the scores of the answer texts and the learning resources. Therefore, it is necessary to score the answer texts and relevant learning resources to measure the matching degree between the answer texts and relevant learning resources and the student respectively.

[0097] In the embodiments of the present invention, after screening out relevant learning resources, optionally, obtain the personalized information of the student from the student information database, and calculate the first interest degree of the student in each of the relevant learning resources; calculate the mastery score of the student in each of the relevant learning resources based on the knowledge points mastery of the student obtained from the student information database; obtain the first matching degree score between each of the relevant learning resources and the question text; perform weighted summation on the first interest degree, the mastery score, and the first matching degree score to obtain the learning resource score of each of the relevant learning resources. According to the personalized information of the student, obtain the second interest degree of the student in the answer text and the second matching degree score between the answer text and the question text, and perform weighted summation on the second interest degree and the second matching degree score to obtain the answer text score. Sort the relevant learning resources and the answer text from high to low according to the learning resource score and the answer text score and display them to the student. For a more detailed process, refer to the previous device part and will not be elaborated here.

[0098] In summary, the personalized intelligent question answering method based on knowledge tracing and knowledge graph in the embodiments of the present invention obtains the question text of the student and obtains the set of retrieved knowledge points of the question text; extracts the set of topic entities from the question text, retrieves the data related to each of the topic entities from the preset knowledge graph and generates an answer text; obtains the relevant learning resources related to the question from the preset learning resource database according to the set of retrieved knowledge points, scores and sorts the relevant learning resources and the answer text respectively according to the personalized information of the student obtained from the student information database, and displays the sorted relevant learning resources and the answer text to the student, which can accurately evaluate the student's knowledge point mastery, and display the answer generated by the knowledge graph and the relevant learning resources personalized according to the student's knowledge point mastery and learning preference characteristics, providing personalized and diversified intelligent question answering services for the student.

[0099] The above describes specific embodiments of the present invention. In some cases, the actions or steps recorded in the embodiments of the present invention can be executed in a different order from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0100] The method of the above embodiment is applied to the corresponding device in the foregoing embodiment and has the beneficial effects of the corresponding device embodiment, which will not be elaborated here.

[0101] Based on the same inventive concept, an embodiment of the present invention further provides an electronic device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the method described in any of the above embodiments is implemented.

[0102] An embodiment of the present invention provides a non-volatile computer storage medium, which stores at least one executable instruction, and the computer executable instruction can execute the method described in any of the above embodiments.

[0103] Figure 3 The following shows a more specific schematic diagram of the hardware structure of the electronic device provided in this embodiment. The device may include: a processor 301, a memory 302, an input / output interface 303, a communication interface 304, and a bus 305. Among them, the processor 301, the memory 302, the input / output interface 303, and the communication interface 304 are communicatively connected to each other inside the device through the bus 305.

[0104] The processor 301 can be implemented in a general way such as a CPU (Central Processing Unit), a microprocessor, an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided by the method embodiments of the present invention.

[0105] The memory 302 can be implemented in forms such as a ROM (Read Only Memory), a RAM (Random Access Memory), a static storage device, a dynamic storage device, etc. The memory 302 can store an operating system and other application programs. When implementing the technical solutions provided by the method embodiments of the present invention through software or firmware, the relevant program codes are stored in the memory 302 and are called and executed by the processor 301.

[0106] The input / output interface 303 is used to connect to an input / output module to implement information input and output. The input / output module can be configured as a component in the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Among them, the input device can include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device can include a display, a speaker, a vibrator, an indicator light, etc.

[0107] The communication interface 304 is used to connect to a communication module (not shown in the figure) to achieve communication and interaction between this device and other devices. The communication module can achieve communication through a wired method (such as USB, network cable, etc.) or through a wireless method (such as mobile network, WIFI, Bluetooth, etc.).

[0108] The bus 305 includes a passage for transmitting information between various components of the device (such as the processor 301, the memory 302, the input / output interface 303, and the communication interface 304).

[0109] It should be noted that although the above device only shows the processor 301, the memory 302, the input / output interface 303, the communication interface 304, and the bus 305, in the specific implementation process, the device may also include other components necessary for normal operation. In addition, those skilled in the art can understand that the above device may also only include the components necessary to implement the solution of the embodiments of the present invention, and does not necessarily include all the components shown in the figure.

[0110] Those of ordinary skill in the art should understand that: the discussion of any of the above embodiments is only exemplary and is not intended to imply that the scope of the present disclosure is limited to these examples; under the concept of the present disclosure, the technical features in the above embodiments or different embodiments can also be combined, and the steps can be implemented in any order, and there are many other variations in different aspects of the embodiments of the present invention as described above, and they are not provided in detail for the sake of brevity.

[0111] This application aims to cover all such substitutions, modifications, and variations that fall within the broad scope of all embodiments. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present invention shall be included within the protection scope of the present disclosure.

Claims

1. A personalized intelligent question answering method based on knowledge tracking and knowledge graph, characterized by: The method comprises: Obtaining the student's question text and obtaining a set of retrieval knowledge points of the question text; Extracting a set of subject entities from the question text, retrieving data related to each of the subject entities from a preset knowledge graph and generating an answer text; Acquire relevant learning resources related to the question from a preset learning resource database according to the retrieved knowledge point set, score and sort the relevant learning resources and the answer text respectively according to the personalized information of the students acquired from the student information database, and present the sorted relevant learning resources and the answer text to the students; Before obtaining the student's question text and the set of retrieval knowledge points of the question text, the method includes: obtaining the student's cognitive state according to the course selected by the student and the corresponding answer data, and saving it in the student information database; applying the K-means clustering algorithm based on the learning style theory according to the behavior data of each student to divide the students into multiple categories, and saving them in the student information database; collecting a variety of learning resources according to the configuration document, and processing the collected multiple learning resources to obtain the relevant text and embedded vector of the learning resources, and storing the embedded vector in the learning resource database; applying the BERT-BiLSTM-CRF model according to the relevant text to extract the triples including the head entity, the relationship, and the tail entity, and storing them in the knowledge graph; The knowledge tracking model is used to evaluate the students’ learning status. The knowledge tracking model uses DKT to apply a recurrent neural network. The formula is as follows: y=RNN(X,W RNN ) K=MLP(y,W M ) Among them, X is the sorted complete answer data; y is the output of RNN; W RNN is the weight of RNN; W M is the weight of the multilayer perceptron MLP, K={η1,η2,…,η n } is the students’ knowledge mastery, that is, the students’ cognitive state, η i It represents the student's mastery of knowledge point i, and its value is between 0 and 1. RNN is a recurrent neural network.

2. The method according to claim 1, characterized in that The step of obtaining the student's cognitive state based on the course selected by the student and the corresponding answer data and saving it in the student information database includes: Push multiple exercises to students based on the courses they choose, and collect the students' corresponding answer data; Acquire the student's historical answer data from the student information database, and concatenate it with the answer data to obtain complete answer data; Applying the deep knowledge tracking model to process the complete answer data, obtaining the students' mastery of each knowledge point, and obtaining the students' cognitive status; The mastery of the knowledge points and the complete answer data are stored in the student information database.

3. The method according to claim 1, characterized in that The step of processing the collected multiple learning resources to obtain relevant texts and embedding vectors of the learning resources includes: Clean the collected learning resources and delete invalid learning resources with problems; Obtain the cleaned relevant texts of each learning resource; Applying the BERT model to generate an embedding vector based on the relevant text; The knowledge point attributes of the learning resource are completed according to the relevant text and the embedding vector.

4. The method according to claim 1, characterized in that: The extracting of triples including head entity, relation, and tail entity by applying BERT-BiLSTM-CRF model according to the relevant text and storing them in the knowledge graph includes: For any sentence in the relevant text, apply the BERT model to obtain the word vector of each word in the sentence; Applying the BiLSTM model to process the word vector of each word to obtain a hidden state vector of each word in the sentence; Input the hidden state vector of each word into the fully connected layer to calculate the probability of each word on all labels; Applying the CRF model to predict the optimal tag sequence of the sentence based on the probability of each word on all tags; The triple <head entity, relationship, tail entity> of the knowledge graph is extracted according to the optimal label sequence and stored in the knowledge graph.

5. The method according to claim 1, characterized in that The step of extracting a set of subject entities from the question text, retrieving data related to each of the subject entities from a preset knowledge graph and generating an answer text includes: Apply the BERT-BiLSTM-CRF model to analyze and process the question text to obtain a label sequence prediction result, and extract each topic entity according to the label sequence prediction result to form a topic entity set; Retrieve each of the subject entities from the knowledge graph, and obtain all relationships existing in each of the retrieved subject entities and the entity nodes pointed to by the relationships; For any relationship, an answer text is generated based on the answer template corresponding to the relationship in the configuration document and the triple <head entity, relationship, tail entity> obtained from the knowledge graph.

6. The method according to claim 1, characterized in that The step of scoring and sorting the relevant learning resources and the answer texts according to the personalized information of the students obtained from the student information database, and presenting the sorted relevant learning resources and the answer texts to the students includes: The personalized information of the students is obtained from the student information database, and the first interest degree of the students in each resource of the related learning resources is calculated; Calculate the students' knowledge mastery from the student information database and score the mastery of each resource in the relevant learning resources; Obtaining a first matching degree score between each resource in the relevant learning resources and the question text; Taking a weighted sum of the first interest level, the mastery level score, and the first matching level score to obtain a learning resource score of each resource in the relevant learning resources; According to the student's personalized information, the student's second interest in the answer text and the second matching degree score between the answer text and the question text are obtained, and the second interest and the second matching degree score are weighted summed to obtain the answer text score; The relevant learning resources and the answer texts are sorted from high to low according to the learning resource scores and the answer text scores and displayed to the students.

7. A personalized intelligent question-answering device based on knowledge tracking and knowledge graph, characterized in that: The device comprises: A question analysis module is used to obtain the student's question text and obtain a set of retrieval knowledge points of the question text; An answer generation module is used to extract a set of subject entities from the question text, retrieve data related to each of the subject entities from a preset knowledge graph and generate an answer text; A personalized question answering module, used to obtain relevant learning resources related to the question from a preset learning resource database according to the retrieved knowledge point set, score and sort the relevant learning resources and the answer text according to the personalized information of the students obtained from the student information database, and present the sorted relevant learning resources and the answer text to the students; The question analysis module, before obtaining the question text of the student and the set of retrieval knowledge points of the question text, includes: obtaining the cognitive state of the student according to the course selected by the student and the corresponding answer data, and saving it in the student information database; applying the K-means clustering algorithm based on the learning style theory according to the behavior data of each student to divide the students into multiple categories, and saving them in the student information database; collecting a variety of learning resources according to the configuration document, and processing the collected multiple learning resources to obtain the relevant text and embedded vector of the learning resources, and storing the embedded vector in the learning resource database; applying the BERT-BiLSTM-CRF model according to the relevant text to extract the triples including the head entity, the relationship, and the tail entity, and storing them in the knowledge graph; The knowledge tracking model is used to evaluate the students’ learning status. The knowledge tracking model uses DKT to apply a recurrent neural network. The formula is as follows: y=RNN(X,W RNN ) K=MLP(y,W M ) Among them, X is the sorted complete answer data; y is the output of RNN; W RNN is the weight of RNN; W M is the weight of the multilayer perceptron MLP, K={η1,η2,…,η n } is the students’ knowledge mastery, that is, the students’ cognitive state, η i It represents the student's mastery of knowledge point i, and its value is between 0 and 1. RNN is a recurrent neural network.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method according to any one of claims 1 to 6 is implemented.

9. A computer storage medium, characterized in that: The storage medium stores at least one executable instruction, and the executable instruction enables the processor to execute the method as described in any one of claims 1-6.

Images