CN114254208A

CN114254208A - Identification method of weak knowledge points and planning method and device of learning path

Info

Publication number: CN114254208A
Application number: CN202111577767.7A
Authority: CN
Inventors: 李睿琪; 沙晶; 王士进
Original assignee: iFlytek Co Ltd
Current assignee: iFlytek Co Ltd
Priority date: 2021-12-22
Filing date: 2021-12-22
Publication date: 2022-03-29

Abstract

The invention provides a weak knowledge point identification method, a learning path planning method and a device, wherein the identification method comprises the following steps: determining a user portrait of a target user based on the test question representation of each test question in the historical answer record of the target user and the historical score condition of the target user corresponding to each test question; determining similar users of the target user based on the user representation of the target user; and determining weak knowledge points of the target user based on the answer score conditions of the similar users corresponding to the knowledge points. The identification method, the planning method and the device provided by the invention avoid the problem of increasing the answering workload of the user, avoid the contingency of the user when answering and evaluating the questions, realize the accurate identification of weak knowledge points of the user, and are beneficial to the user to carry out targeted learning.

Description

Identification method of weak knowledge points and planning method and device of learning path

Technical Field

The invention relates to the technical field of computers, in particular to a weak knowledge point identification method, a learning path planning method and a learning path planning device.

Background

With the coming of the internet and big data era, the progress of education informatization is deepened continuously, and online education is brought forward. Different from the traditional live-broadcast lesson mode, most AI (Artificial Intelligence) interactive lessons do not need the lessee to participate in the whole course, dynamically use the modes of video teaching, problem training and the like to guide students to participate in the interaction, lead the students to efficiently learn the difficulty of the lessons, and cultivate the ability of the students to study consciously.

An existing AI interactive course generally sets an evaluation link in an interactive teaching process, and weak knowledge points of students are diagnosed according to answer expressions of the students in the evaluation link. However, this kind of diagnosis method is too simple to avoid the contingency of students in answering and evaluating questions, and it is difficult to diagnose the real weak points of students.

Disclosure of Invention

The invention provides a weak knowledge point identification method, a learning path planning method and a learning path planning device, which are used for solving the defect of inaccurate identification of weak knowledge points in the prior art and realizing accurate identification of weak knowledge points of users.

The invention provides a method for identifying weak knowledge points, which comprises the following steps:

determining a user portrait of a target user based on the test question representation of each test question in the historical answer record of the target user and the historical score condition of the target user corresponding to each test question;

determining similar users of the target user based on the user representation of the target user;

and determining weak knowledge points of the target user based on the answer score conditions of the similar users corresponding to the knowledge points.

According to the identification method of weak knowledge points provided by the invention, the user portrait of the target user is determined based on the test question representation of each test question in the historical answer record of the target user and the historical score condition corresponding to each test question of the target user, and the identification method comprises the following steps:

based on the test question representation of each test question, the historical score condition of the target user corresponding to each test question and the user portrait mapping relation, constructing a user portrait of the target user to obtain the user portrait of the target user;

the user portrait mapping relationship is determined based on consistency between a first user portrait corresponding to a first answer record of a sample user and a second user portrait corresponding to a second answer record of the sample user, and difference between the first user portrait and a negative user portrait corresponding to a negative answer record of a negative user, the second answer record corresponding to the same test question as the negative answer record.

According to the identification method of the weak knowledge points, provided by the invention, each knowledge point is determined based on the following steps:

determining similar test questions similar to the teaching video based on the similarity between the video characteristics of the teaching video and the test question characteristics of each candidate test question, wherein the video characteristics and the test question characteristics are in the same semantic space;

and determining each knowledge point contained in the teaching video based on the test question knowledge points of the similar test questions.

According to the identification method of the weak knowledge points, provided by the invention, the test question knowledge points of the similar test questions are determined based on the following steps:

extracting the characteristics of the similar test questions to obtain first text characteristics of the similar test questions;

coding the similar test questions based on the similarity between each word and each candidate knowledge point in the similar test questions to obtain second text characteristics of the similar test questions;

and predicting knowledge points based on the first text characteristics and the second text characteristics of the similar test questions to obtain the test question knowledge points of the similar test questions.

According to the identification method of the weak knowledge points, provided by the invention, the video characteristics of the teaching video are determined based on the following steps:

extracting text features, image features and audio features of the teaching video;

and performing feature fusion on the text feature, the image feature and the audio feature of the teaching video to obtain the video feature of the teaching video.

The invention also provides a method for planning the learning path, which comprises the following steps:

determining weak knowledge points of the target user based on the answer score conditions of the similar users corresponding to the knowledge points;

and determining a learning path corresponding to the target user based on the weak knowledge points of the target user.

According to the planning method of the learning path provided by the invention, the step of determining the learning path corresponding to the target user based on the weak knowledge points of the target user comprises the following steps:

and determining the learning path based on the video slices and the test questions corresponding to the weak knowledge points and the attribute information of the weak knowledge points.

The invention also provides a device for identifying weak knowledge points, which comprises:

the user portrait determining module is used for determining a user portrait of a target user based on the test question representation of each test question in the historical answer record of the target user and the historical score condition of the target user corresponding to each test question;

a similar user determination module for determining a similar user of the target user based on the user representation of the target user;

and the knowledge point determining module is used for determining weak knowledge points of the target user based on the answer score conditions of the similar users corresponding to the knowledge points.

The invention also provides a planning device for learning the path, which comprises:

a knowledge point determining module, configured to determine weak knowledge points of the target user based on answer score conditions corresponding to the knowledge points of the similar users;

and the learning path planning module is used for determining a learning path corresponding to the target user based on the weak knowledge points of the target user.

The invention further provides an electronic device, which includes a memory, a processor and a computer program stored in the memory and capable of running on the processor, wherein the processor implements the method for identifying weak knowledge points or the method for planning learned paths as described above when executing the program.

The present invention also provides a non-transitory computer-readable storage medium having stored thereon a computer program which, when being executed by a processor, implements the method for identifying weak knowledge points as described in any of the above, or the steps of the method for planning a learned path as described in any of the above.

The identification method of weak knowledge points and the planning method and the device of learning paths provided by the invention can obtain the user portrait capable of representing the learning ability of the target user through the test question representation of each test question in the historical answer record of the target user and the historical score condition of each test question corresponding to the target user, determine the similar user of the target user based on the user portrait, and then determine the weak knowledge points of the target user based on the historical answer representation of the similar user corresponding to each knowledge point, thereby realizing the identification of the weak knowledge points of the target user without answering the target user at each knowledge point, avoiding the problem of increasing the answering workload of the user, avoiding the contingency of the user when answering and testing and evaluating the questions, realizing the accurate identification of the weak knowledge points of the user, being beneficial to the user to carry out targeted learning, being also beneficial to carry out personalized learning resource recommendation aiming at the weak knowledge points subsequently, the interpretability and acceptability of the learning resource recommendation are improved.

Drawings

In order to more clearly illustrate the technical solutions of the present invention or the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.

FIG. 1 is a schematic flow chart of a weak knowledge point identification method provided by the present invention;

FIG. 2 is a schematic flow chart of a method for determining knowledge points according to the present invention;

FIG. 3 is a flow chart of the method for determining knowledge points of test questions provided by the present invention;

FIG. 4 is a block diagram of a model for predicting knowledge points of test questions according to the present invention;

FIG. 5 is a block diagram of a user representation model provided by the present invention;

FIG. 6 is a flow chart illustrating a method for determining video characteristics according to the present invention;

FIG. 7 is a flow chart illustrating a method for determining video features and test question features according to the present invention;

FIG. 8 is a schematic flow chart of a method for planning a learned path according to the present invention;

FIG. 9 is a flowchart illustrating an adaptive scheduling method for AI interactive courses according to the present invention;

FIG. 10 is a schematic structural diagram of a weak knowledge point recognition apparatus provided by the present invention;

FIG. 11 is a schematic structural diagram of a device for planning a learned route according to the present invention;

fig. 12 is a schematic structural diagram of an electronic device provided in the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is obvious that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

Online education is generally divided into two major categories, live and AI interactive lessons. Different from the traditional live broadcast class mode, AI interactive class is mostly shown through the form of recorded broadcast, does not need the teacher who gives lessons to participate in whole journey, utilizes artificial intelligence and big data technology, for its personalized study scheme of customization according to student's knowledge leak, guides the student to carry out video study and exercise training, lets the student high-efficient study course difficult point to cultivate the ability of student's conscious study. The AI interactive course is characterized in that the system leads teaching, teachers assist in answering questions, and high dependence on teachers and resources is reduced; meanwhile, the interactive learning makes the teaching process more vivid, the learning behavior of the students is not limited by time and space, and the subjective activity of the students can be stimulated.

An existing AI interactive course generally sets an evaluation link in an interactive teaching process, and weak knowledge points of students are diagnosed according to answer expressions of the students in the evaluation link. However, this kind of diagnosis method is too simple to avoid the contingency of students in answering and evaluating questions, and it is difficult to diagnose the real weak points of students. In addition, the total answer amount of the students is increased by the arrangement of the evaluation links, and the efficient learning of the students is not facilitated.

In order to solve the problems, the invention provides a method for identifying weak knowledge points. Fig. 1 is a schematic flow chart of a method for identifying weak knowledge points, which is provided by the present invention, and as shown in fig. 1, the method includes:

step 110, determining a user portrait of the target user based on the test question representation of each test question in the historical answer record of the target user and the historical score condition of the target user corresponding to each test question;

step 120, determining similar users of the target user based on the user portrait of the target user;

and step 130, determining weak knowledge points of the target user based on the historical answer score conditions corresponding to the knowledge points of the similar users.

Specifically, the target user refers to a user to be subjected to weak knowledge point identification, and the historical answer record is used for describing relevant information of the target user on completed test questions, such as the answer result, test question score, test question text, test question attribute and the like of the target user, and may be an answer record of one test or an answer record of multiple tests. The test question representation is a coded representation of the test question, and the test question representation can contain semantic information of each participle in each test question and can also contain context information of each participle. The history score may be, for example, a history score rate, a history average score, or a history score total value of the target user for answering each test question, and this is not particularly limited in the embodiment of the present invention.

According to the test question representation of each test question in the historical answer record of the target user and the historical score condition of the target user corresponding to each test question, the user portrait of the target user is determined, the user portrait can represent the learning capacity of the target user, and then according to the user portrait of the target user, similar users of the target user are determined from candidate users, so that users with the learning capacity level similar to that of the target user can be obtained, and the candidate users are users possibly similar to the learning capacity of the target user.

Here, the method of determining the user image of the target user may be specifically obtained by simple operations such as multiplication, division, weighted sum, and averaging of the test question representation of each test question and the history score condition of the target user corresponding to each test question, may be obtained by inputting the test question representation of each test question and the history score condition of the target user corresponding to each test question into the user image model, and then outputting the user image of the target user by the user image model, or may be obtained by combining the test question representation of each test question and other information such as the history score condition of the target user corresponding to each test question and the average score rate of each test question, and this is not particularly limited in the embodiments of the present invention.

The determination mode of the similar users may specifically be determined according to the similarity between the user image of the target user and the user images of the candidate users, or may be determined by first screening a part of users from the candidate users according to the difference between the target user and the user images of the candidate users, and then further determining the similar users according to the similarity between the user image of the target user and the user images of the screened part of users. For example, the candidate users include users A, B and C, the difference between the score of the target user and the score of user a is not within a preset range by calculating the difference between the score of the target user and the score of each candidate user, and at this time, the similarity between the user images of users B and C and the user image of the target user only needs to be calculated, and if the similarity between the user image of the target user and the user image of user B meets a preset condition, it indicates that user B is a similar user of the target user.

Based on this, because the learning abilities of the similar user and the target user are similar, the estimated answer score condition of the target user corresponding to each knowledge point can be estimated according to the historical answer score condition of the similar user corresponding to each knowledge point, and then, the weak knowledge point of the target user can be determined according to the estimated answer score condition of the target user corresponding to each knowledge point, for example, according to the historical answer record of the similar user under a certain knowledge point, the historical answer score or the total score value of the similar user corresponding to the knowledge point is higher, that is, the estimated answer score or the total score value of the target user corresponding to the knowledge point is higher, so that the knowledge point is not the weak knowledge point of the target user; on the contrary, if the historical answer score or the total score value corresponding to a certain knowledge point by the similar user is lower, the knowledge point can be determined to belong to the weak knowledge point of the target user. Here, each knowledge point may be a knowledge point in a knowledge system constructed in advance, or may be a knowledge point included in a resource such as a teaching image or a video, and this is not particularly limited in the embodiment of the present invention.

It should be noted that, since it needs to be applied to the historical answer score condition of the similar user corresponding to each knowledge point in step 130, for each knowledge point, the user having the historical answer record under the knowledge point may be selected as a candidate user, or a part of the users having the historical answer record under the knowledge point may be selected as candidate users according to the information of the age, the grade, and the like of the user. The determined weak knowledge points of the target user can be applied to weak knowledge point identification scenes in various fields, for example, in the field of traditional education, the target user can quickly know subject knowledge points with low self-mastering degree by identifying the weak subject knowledge points of the target user, the learning capacity is improved in a targeted manner, and resources such as course contents and test questions corresponding to the weak subject knowledge points can be pushed to the target user so as to strengthen learning of the weak subject knowledge points by the target user.

The method provided by the embodiment of the invention obtains the user portrait capable of representing the learning ability of the target user through the test question representation of each test question in the history answer record of the target user and the history scoring condition of the target user corresponding to each test question, determines the similar user of the target user based on the user portrait, and determines the weak knowledge point of the target user based on the history answer representation of the similar user corresponding to each knowledge point, thereby realizing the identification of the weak knowledge point of the target user without answering the target user at each knowledge point, avoiding the problem of increasing the answering workload of the user, avoiding the contingency of the user when testing and evaluating the questions, realizing the accurate identification of the weak knowledge point of the user, being beneficial to the targeted learning of the user, being also beneficial to the personalized learning resource recommendation of the weak knowledge point subsequently, the interpretability and acceptability of the learning resource recommendation are improved.

Based on any of the above embodiments, step 110 includes:

the user portrait mapping relationship is determined based on a correspondence between a first user portrait corresponding to a first answer record of the sample user and a second user portrait corresponding to a second answer record of the sample user, and a difference between the first user portrait and a negative user portrait corresponding to a negative answer record of the negative user, the second answer record corresponding to the same test question as the negative answer record.

Specifically, the user portrait construction may be performed on the target user according to the test question representation of each test question in the history answer record of the target user, the history score condition corresponding to each test question of the target user, and the mapping relationship between the sample user and the user portrait thereof, that is, the user portrait mapping relationship, to obtain the user portrait of the target user.

In order to realize accurate user portrait construction, the embodiment of the invention is directed to a determination process of a user portrait mapping relation, namely a training process of a user portrait model or a mining process of user portrait construction rules, and applies a first answer record of a sample user, a second answer record of the sample user and a negative answer record of the negative answer user, wherein the second answer record corresponds to the same test question as the negative answer record. By adjusting parameters of the user portrait model or a representation mode of a user portrait construction rule, a first user portrait and a second user portrait, namely user portraits constructed on the basis of different answer records of the same user, can be made to be similar as much as possible, and the first user portrait and a negative user portrait, namely user portraits constructed on the basis of answer records of different users, can be made to be different as much as possible, so that the mapping relation obtained by the method can be more reliable after the method is applied to user portrait construction.

For example, the user profile model may be pre-trained by: firstly, collecting historical answer records of sample users, and selecting a first answer record and a second answer record which are associated with the historical answer records, and a negative example user corresponding to the sample user, wherein the negative example answer records of the negative example user correspond to the same examination questions as the second answer records, for example, considering that the learning ability of the same user before and after the examination is relatively stable, 5 examination records of the sample user before a certain scale examination (such as a monthly exam, a joint exam, an end-of-term exam and the like) can be used as the first answer records, the record of the certain scale examination can be used as the second answer records, the same class of the sample user can be used as the negative example user, and the record of the negative example user in the certain scale examination can be used as the negative example answer records; then, training an initial model by taking the consistency between a first user portrait corresponding to the first answer record of the sample user and a second user portrait corresponding to the second answer record of the sample user and the difference between the first user portrait and a negative user portrait corresponding to the negative answer record of the negative user as targets, so as to obtain a user portrait model, which can be realized by a loss function adopting the following formula, for example:

where pre _ self represents a first user representation of a sample user, next _ self represents a second user representation of the sample user, next _ else _ i represents a negative user representation of an ith negative user, n represents the number of negative users, and cosine represents cosine similarity.

In the training process, the user portrait model can respectively obtain a first user portrait of the sample user, a second user portrait of the sample user and a negative user portrait of the negative user based on a first answer record of the sample user, a second answer record of the sample user and a negative answer record of the negative user, and the consistency between the first user portrait of the sample user and the second user portrait and the difference between the first user portrait of the sample user and the negative user portrait of the negative user are taken as training targets, so that the user portrait model can accurately represent the learning ability of the user and has stronger robustness.

On the basis, the test question representation of each test question in the historical answer record of the target user and the historical score condition of the target user corresponding to each test question are input into the user portrait model, the user portrait model constructs the user portrait of the target user, and the accurate user portrait of the target user can be obtained.

Based on any of the above embodiments, fig. 2 is a schematic flow chart of the determination method of knowledge points provided by the present invention, and as shown in fig. 2, the knowledge points are determined based on the following steps:

step 210, determining similar test questions similar to the teaching video based on the similarity between the video characteristics of the teaching video and the test question characteristics of each candidate test question, wherein the video characteristics and the test question characteristics are in the same semantic space;

step 220, determining knowledge points contained in the teaching video based on the test question knowledge points of the similar test questions.

Specifically, the teaching video may be a blackboard writing video, a ppt video, or the like directly recorded in the course of the teacher giving a lesson, or may be a teaching demonstration video or the like prepared in advance. The instructional video may be represented as a sequence of image frames arranged from front to back in a temporal order. The candidate test questions may be test questions that may be adapted to the teaching video, and the candidate test questions may be obtained directly from the mass resource question bank, or obtained by screening test questions from the mass resource question bank according to subjects to which the teaching video belongs or other information.

In order to screen out similar test questions adapted to the teaching video from the candidate test questions, feature extraction can be performed on the teaching video and the candidate test questions respectively, so that video features of the teaching video and test question features of the candidate test questions are obtained. The feature extraction for the teaching video may be implemented by a pre-trained coding module or a pre-set coding rule, because the teaching video relates to multiple different resource modalities, the different resource modalities may be coded in corresponding modalities respectively, and then coding results of the different resource modalities are integrated to obtain video features, where the integration may be direct splicing, or weighted fusion, or feature fusion based on an attention mechanism, and this is not specifically limited in the embodiment of the present invention. Similarly, the feature extraction for each candidate test question may be implemented by a pre-trained coding module, or may also be implemented by a pre-set coding rule, which is not specifically limited in this embodiment of the present invention.

In order to calculate the similarity between the video features of the teaching video and the test question features of the candidate test questions and mine semantic association relations contained between different resources, the video features and the test question features obtained by feature extraction need to be in the same semantic space, namely the video features and the test question features are in the same vector space. The video features and the test question features are in the same semantic space, and the video features and the test question features can be specifically realized by setting a matched coding module or coding rule for feature extraction of the teaching video and feature extraction of each candidate test question in advance. On the basis, the similarity between the video characteristics of the teaching video and the test question characteristics of the candidate test questions can be calculated, and the similar test questions similar to the teaching video are selected from the candidate test questions according to the calculation result. Then, according to the test question knowledge points of the similar test questions, all the knowledge points contained in the teaching video can be determined.

Here, the number of the similar test questions may be one or more, the method of selecting the similar test questions based on the similarity may be to sort the similarity between the video features and the features of the test questions in order from high to low, select a preset number of candidate test questions with the highest similarity as the similar test questions, or determine the candidate test questions with the similarity higher than a preset threshold as the similar test questions, where the preset threshold may be preset, and may be, for example, 75% or 80%. The test question knowledge points, that is, the knowledge points examined by the candidate test questions, may be obtained by manual labeling or by prediction of the knowledge points, and the embodiment of the present invention is not particularly limited in this respect.

According to the method provided by the embodiment of the invention, similar test questions similar to the teaching video are obtained by carrying out feature matching on the basis of the video features and the test question features in the same semantic space, and then each knowledge point contained in the teaching video is determined on the basis of the test question knowledge points of the similar test questions, so that the knowledge point identification of the teaching video can be realized without manually marking the knowledge point labels related to the teaching video and manually associating the mapping relation between the teaching video and the candidate test questions, the time and the labor are saved, and the accuracy and the efficiency of the knowledge point identification of the teaching video are improved.

Based on any of the above embodiments, fig. 3 is a schematic flow chart of the method for determining test question knowledge points provided by the present invention, and as shown in fig. 3, the test question knowledge points of similar test questions are determined based on the following steps:

step 310, extracting the characteristics of the similar test questions to obtain first text characteristics of the similar test questions;

step 320, coding the similar test questions based on the similarity between each word in the similar test questions and each candidate knowledge point to obtain second text characteristics of the similar test questions;

and step 330, predicting knowledge points based on the first text characteristics and the second text characteristics of the similar test questions to obtain test question knowledge points of the similar test questions.

Specifically, in consideration of the existing knowledge point labeling of the test questions, the teaching and research personnel are required to label a proper knowledge point for each test question, and the manual labeling mode highly depends on the teaching and research personnel and the professional level thereof, is time-consuming and labor-consuming, has too high labor cost, and is not beneficial to large-scale popularization. In addition, the manual labeling mode has strong subjectivity, and the labeling results of different people on the same test question are easy to be inconsistent.

Therefore, the embodiment of the invention firstly extracts the characteristics of the similar test questions to obtain the first text characteristics of the similar test questions, codes the similar test questions according to the similarity between each word in the similar test questions and each candidate knowledge point to obtain the second text characteristics of the similar test questions, and then predicts the knowledge points by combining the first text characteristics and the second text characteristics of the similar test questions, thereby obtaining the accurate test question knowledge points of the similar test questions without manual participation.

Here, the first text feature may be obtained by performing feature extraction according to word vectors of words in the similar test questions to obtain the first text feature, or may be obtained by performing feature extraction according to text vectors of the similar test questions to obtain the first text feature. The candidate knowledge points may be knowledge points that may be matched with the similar test questions, and the candidate knowledge points may be acquired from a pre-constructed knowledge system.

It should be noted that, in the process of obtaining the second text feature by encoding, the similarity between each word in the similar test question and each candidate knowledge point is applied, so that the contribution of each word to the classification of the similar test question can be analyzed by introducing the semantic information of the label, i.e., the candidate knowledge point, so that the second text feature can contribute to the classification of the similar test question, and the knowledge point prediction is performed by combining the first text feature obtained by feature extraction, so that the prediction accuracy of the test question knowledge point can be greatly improved.

Based on any of the above embodiments, fig. 4 is a schematic diagram of a framework of a test question knowledge point prediction model provided by the present invention, and as shown in fig. 4, test question knowledge points of similar test questions can be obtained through prediction of the test question knowledge point prediction model:

s1, splicing the subject surface, options, answers and analysis of the similar test questions together to serve as input, and mapping each word in the similar test questions into a word vector by using a word to vector model pre-trained by a large amount of word segmentation corpora, so as to obtain a word vector sequence (namely, Embedding in FIG. 4).

S2, on one hand, inputting the word vector sequence into a textCNN (relational Networks for Session Classification) (namely convolution one, convolution two and convolution three in FIG. 4) for feature extraction, and obtaining the first text feature corresponding to the similar test question.

And S3, on the other hand, calculating a cosine similarity matrix (dimension is V multiplied by L, V is the size of a word list, and L is the total number of the candidate knowledge points) of each word and each candidate knowledge point in the similar test question in advance. And inputting the cosine similarity matrix into an LEAM (Label Embedding protected Model, supervised text classification Model), and coding the similar test questions by the LEAM based on the cosine similarity matrix to obtain second text characteristics of the similar test questions. Here, the cosine similarity matrix is regarded as a similarity vector sequence of each word and each candidate knowledge point, the maximum support degree of each word by the candidate knowledge point under the n-gram viewing angle is further calculated in a convolution mode, the maximum support degree is used as the contribution degree of each word to the classification of similar test questions, and the contribution degree is used for carrying out weighted summation on the word vectors of each word, so that the second text characteristic after the word weight is corrected is obtained.

And S4, splicing the two parts of text features obtained in S2 and S3 together by using a full connection layer, and calculating the final classification probability to finally obtain the test question knowledge points of similar test questions.

Based on any embodiment, the user portrait model is used for describing an implicit portrait of the user, so as to express the learning ability of the user, that is, the learning situation vector of the user, and the closer the learning situation vector distance is, the higher the user similarity is. Accordingly, a user representation of the target user may be obtained in step 110 based on the user representation model. In addition, considering that the end-of-term examination can reflect the learning ability of the user better than other examinations, the embodiment of the invention can train the user portrait model by using the first answer record (pre) of the sample user before the end-of-term examination, the second answer record (next) of the sample user in the end-of-term examination and the negative answer record (next-else) of the negative user in the end-of-term examination.

Fig. 5 is a schematic diagram of a framework of a user portrait model according to the present invention, as shown in fig. 5, in a training process, the framework is divided into two parts, namely, a pre-end-of-term examination and an end-of-term examination, and finally, in a matching layer, a loss of the model is determined based on a hidden layer representation obtained by the two parts and a loss function, and training is performed with the loss as a constraint, and the following description will be made by taking the part before the end-of-term examination as an example:

1. the input layer receives first answer records of sample users, each input node represents that ht points of the test question representation of each test question in the first answer records are multiplied by (si-di)/stdi, si represents the score rate of the sample users on the test question, di represents the average score rate of the test question, stdi represents the standard deviation of the score rate of the test question, and the test question representation can be obtained by inputting the test question stems into a test question knowledge point prediction model and extracting feature representations of a full-connection layer from the test question stems;

2. the Mean layer averages all input nodes of the input layer;

3. the Linear layer performs Linear transformation on the output of the Mean layer;

4. the LayNorm layer standardizes the output of the Linear layer;

5. the Sigmoid layer performs Sigmoid nonlinear transformation on the output of the LayNorm layer;

6. the "hidden layer representation" of the user is finally obtained.

In practical application, based on the historical answer records of the target user, the user portrait of the target user can be expressed in the following three ways:

1. the perceptron model: using the "hidden layer representation" (pre part/next part) in FIG. 5 as a user representation of the target user;

2. hidden vector model: using the "Mean" (pre section) in fig. 5 as the user representation of the target user;

3. average score model: and directly using the pre-examination average score in the historical answer record of the target user as the user portrait of the target user.

Similar users are searched based on user figures, cosine similarity of vectors of two users is needed to measure user similarity in the mode 1 and the mode 2, and the absolute value of the difference value of the average scoring rates of the two users is directly used to measure the user similarity in the mode 3. In practical application, an average score (avg) filtering step can be added, namely, a similar user range is limited through an average score model, and then similar users are searched by using 'hidden layer representation' of a perceptron model, so that the calculation efficiency is improved, the interpretability of a recommendation result is enhanced, and the use experience of the users is optimized. After the similar users are obtained, whether each knowledge point is a weak knowledge point of the target user or not can be diagnosed by taking the average score of the real historical answer records of the similar users under the knowledge points as a basis, and learning resource recommendation is carried out on the target user based on the weak knowledge points.

Based on any of the above embodiments, fig. 6 is a schematic flow chart of the method for determining video features provided by the present invention, and as shown in fig. 6, the video features of the teaching video are determined based on the following steps:

step 610, extracting text features, image features and audio features of the teaching video;

and step 620, performing feature fusion on the text feature, the image feature and the audio feature of the teaching video to obtain the video feature of the teaching video.

Specifically, the teaching video may include multi-modal information data such as teacher explanation audio data, image data such as course blackboard writing and courseware, and course summary text data. For the multi-modal condition, feature extraction suitable for various modalities can be performed on various modality information data in the teaching video respectively:

for audio data, when feature extraction of an audio mode is performed, a suitable depth model may be selected according to an input form of the audio mode to perform feature extraction, for example, after frame division and windowing are performed on one-dimensional audio data, acoustic features of each frame, such as Mel Frequency Cepstrum Coefficient (MFCC) features or Perceptual Linear Prediction (PLP) features, are extracted through Fast Fourier Transform (FFT), or global feature representation is extracted through a Multi-Layer Perceptron (MLP), and for example, a sound spectrum diagram of two-dimensional audio data is used as image data, and spatial feature extraction is performed by using a Convolutional Neural Network (CNN).

For the image data, considering that although the image sequence of the video is richer than information contained in a single image, redundant information in the image sequence is excessive, a plurality of key frame images can be extracted from the image sequence of the teaching video by a key frame extraction method to perform feature extraction in an image mode. Here, the key frame extraction method may be selected according to different video types, for example, a shot-based key frame extraction algorithm: dividing a teaching video according to shot changes, and then selecting a first frame and a last frame in each shot of the teaching video as key frame images; the video clustering-based method comprises the following steps: the teaching video is divided into a plurality of clusters through clustering, and corresponding video frames are selected from each cluster to serve as key frame images. Further, when feature extraction of an image modality is performed on a key frame image, a convolutional neural network can be applied to extract image features with rotational-translational scaling invariance. It can be understood that, since each key frame image has its own features that can be distinguished from other key frame images, the image features obtained by feature extraction of each key frame image are also different from those obtained by other key frame images.

For text data, when feature extraction of a text mode is performed, each Word in the text data can be mapped into a Word vector through a pre-trained Word embedding model, such as Word2Vec, GloVe, Bert, and the like, and then, a text feature corresponding to the text data is obtained through each Word vector. Here, the text feature of the text data is obtained by each word vector, each word vector may be directly integrated into a vector sequence as the text feature, or the vector sequence integrated by each word vector may be input into a Long Short-Term Memory Network (LSTM) or a Recurrent Neural Network (RNN) to obtain a semantic representation of each word, and then the text feature is obtained by integration. In the process, considering that the words are the finest granularity formed by the natural language text, the words form sentences, the sentences form paragraphs, chapters and documents, when the text characteristics of the text data are integrated, the words and the vectors can be integrated into sentence vectors by the logic, and the sentence vectors are integrated into paragraph vectors, chapter vectors, document vectors and the like until the granularity of the text data is achieved.

Through the multi-modal feature extraction step, independent semantic codes of each modality, that is, text features, image features and audio features corresponding to the teaching video are obtained, and then the text features, the image features and the audio features can be subjected to feature fusion, so as to obtain video features of the teaching video.

According to the method provided by the embodiment of the invention, the video characteristics of the teaching video are obtained by respectively extracting the characteristics of the information data of various modes in the teaching video and then performing characteristic fusion on the characteristics of various modes obtained by characteristic extraction, so that more accurate and reliable similar test questions can be obtained when the characteristics are matched based on the video characteristics and the test question characteristics in the follow-up process.

Based on any of the above embodiments, fig. 7 is a schematic flowchart of the method for determining video features and test question features provided by the present invention, as shown in fig. 7, for a multi-modal case, in order to obtain video features, features of each modal data may be first mapped into a uniform multi-modal embedding space, for example, feature mapping may be performed in a full-connected manner, and a calculation formula is as follows:

wherein f is_T，f_I，f_SRespectively obtaining X through feature mapping for the features of Text data (Text), Image data (Image) and audio data (Sound) obtained by feature extraction_T，X_I，X_SMapping features for text data, image data and audio data, respectively. W_T，W_I，W_SAnd b_T，b_I，b_SEmbedded matrices and offset vectors for corresponding text, images and audio.

In addition, each candidate test question may include multi-modal information data such as text data and image data, and similar feature extraction and feature mapping may be performed for each candidate test question according to a processing method of a multi-modal teaching video, so as to obtain a mapping feature X of text and image data in the candidate test question_T′，X_I′。

Then, considering that semantic association exists between the multi-modal video resources and the multi-modal test question resources, semantic alignment and information fusion need to be carried out through multi-modal data fusion, and on the other hand, the video resources and the test question resources need to be deeply represented in a unified manner so as to conveniently mine the association relation contained in the resources in a knowledge level, so that the embodiment of the invention can utilize Deep Factorization Machines (Deep Factorization Machines, Deep fm), Deep cross networks (Deep frames)&Cross Network, DCN), etc. model pair X_T，X_I，X_SAnd X_T′，X_I' equal characteristic data are fused, the video characteristics and the test question characteristics are mapped to the same vector space, the associated knowledge modeling of different types of education resources is completed, and therefore the video characteristics X in the same semantic space are obtained_FAnd test question feature X_Q。

Based on the above, the test question features of the candidate test questions can be obtained and stored in the vearch of the distributed vector search system. The vearch can be used for storing and calculating massive feature vectors, has a function of fast vector retrieval, and can guarantee millisecond-level response. After the video features of the current teaching video are obtained, vector retrieval can be carried out by using the vearch based on the video features to obtain a plurality of test question features similar to the video features, and the test questions corresponding to the test question features can be determined to be similar test questions. Then, test question knowledge points of similar test questions are obtained through knowledge point prediction, and a plurality of related knowledge points contained in the teaching video can be obtained.

Based on any one of the embodiments, considering that the existing AI interactive lesson generally sets an evaluation link in the interactive teaching process, and diagnosing weak knowledge points of students according to the answer performance of the students in the evaluation link. The diagnosis method is too simple, the contingency of students in answering and evaluating questions cannot be avoided, the real weak points of the students cannot be diagnosed, so that proper learning paths cannot be matched for the students, the total answer amount of the students is increased due to the arrangement of the evaluating links, and the efficient learning of the students is not facilitated.

Aiming at the problems, the invention provides a method for planning a learning path. Fig. 8 is a schematic flow chart of a method for planning a learned path according to the present invention, and as shown in fig. 8, the method includes:

step 810, determining a user portrait of the target user based on the test question representation of each test question in the historical answer record of the target user and the historical score condition corresponding to each test question of the target user;

step 820, determining similar users of the target user based on the user portrait of the target user;

step 830, determining weak knowledge points of the target user based on the answer score conditions of the similar users corresponding to the knowledge points;

and step 840, determining a learning path corresponding to the target user based on the weak knowledge points of the target user.

The determination mode of the similar users may specifically be determined according to the similarity between the user image of the target user and the user images of the candidate users, or may be determined by first screening a part of users from the candidate users according to the difference between the target user and the user images of the candidate users, and then further determining the similar users according to the similarity between the user image of the target user and the user images of the screened part of users. For example, the candidate users include users A, B and C, and the difference between the score of the target user and the score of user a is found to be out of the preset range by calculating the difference between the score of the target user and the score of each candidate user, at this time, the similarity between the user images of users B and C and the user image of the target user only needs to be calculated, and if the similarity between the user image of the target user and the user image of user B meets the preset condition, it indicates that user B is a similar user of the target user.

It should be noted that, since it needs to be applied to the historical answer score condition of the similar user corresponding to each knowledge point in step 830, for each knowledge point, the user having the historical answer record under the knowledge point may be selected as a candidate user, and a part of the users may be selected as candidate users from the users having the historical answer record under the knowledge point according to the information of the age, the grade, and the like of the users.

After weak knowledge points of the target user are obtained, learning paths corresponding to the target user can be organized according to the weak knowledge points, personalized learning resources such as course images, videos and test questions are pushed, the target user is guided to selectively complete learning according to the learning paths, and the learning of the target user is enabled to be simple and effective.

The method provided by the embodiment of the invention can realize the test question representation of each test question in the historical answer record based on the target user, and the historical score condition of the target user corresponding to each test question, and determines the similar user of the target user based on the obtained user figure, and then determines the weak knowledge point of the target user based on the historical answer expression of the similar user corresponding to each knowledge point, thereby avoiding the problem of increasing the work load of answering questions of the user, avoiding the contingency of the user when answering and evaluating the questions, realizing the accurate identification of weak knowledge points of the user, on the basis, the learning path corresponding to the target user is determined based on the weak knowledge points of the target user, personalized learning resource recommendation can be performed aiming at the weak knowledge points, personalized teaching is achieved, the learning effect and efficiency of the target user are further improved, and meanwhile, the interpretability and acceptability of the learning resource recommendation are also improved.

Based on any of the above embodiments, step 840 comprises:

and determining a learning path based on the video slices and the test questions corresponding to the weak knowledge points and the attribute information of the weak knowledge points.

Specifically, it is considered that in the interactive teaching process of the existing AI interactive lesson, the similar question group with the wrong answer in the evaluation link is usually recommended for students to train so as to achieve the purpose of strengthening weak knowledge learning, however, in this way, only the similar question group with the wrong answer in the evaluation link is recommended, the front-back dependency relationship between different knowledge points is often ignored, the difficulty progressive relationship between different exercise questions of the same knowledge point is ignored, the learning difficulty of the students is increased due to the lack of the progressive learning process, and the participation degree of the students is reduced.

In contrast, according to the weak knowledge points, the embodiment of the invention organizes the learning path corresponding to the target user according to the video slices, the practice problems, the evaluation problems and other test problems in the teaching video and the attribute information of the weak knowledge points, so that the target user is guided to selectively complete learning according to the learning path, and the learning of the target user becomes simple and effective. Here, the attribute information may be, for example, an exposure rate of a knowledge point in a scale test, a difficulty level of the knowledge point, a teaching order of the knowledge point, and related information with other knowledge points, and the exposure rate of the knowledge point may be determined based on a test frequency and a score value of the knowledge point.

According to the attribute information of the weak knowledge points, the learning sequence of each knowledge point in the learning path can be organized, for example, A is the weak knowledge point and is an integrated knowledge point, so that when the learning path is organized, the resources corresponding to the knowledge points associated with A can be placed in front of the resources corresponding to A for learning, for example, B and C are both weak knowledge points, and B has a lower difficulty level than C, so that when the learning path is organized, the resources corresponding to the knowledge points B can be learned first, and then the resources corresponding to the knowledge points C can be learned.

Further, after the target user enters the learning path, video slices, practice problem groups, evaluation problems and the like can be dynamically put in according to the real answering situation of the target user at each knowledge point, for example, if the score rate of answering by the target user at a certain knowledge point is higher, resources such as videos, images, test problems and the like at the knowledge point can be properly deleted, and test problems with increased difficulty can be recommended to be exercised, so that the learning path is dynamically updated, ineffective exercises are reduced, the learning efficiency is improved, and the dynamic planning of the learning path is realized.

Based on any of the above embodiments, considering that the existing AI interactive lessons generally make lectures and record courses for manual work, the association between teaching videos and knowledge systems needs to be completed by the reference of teaching and research experts, and the teaching and research experts arrange appropriate supporting exercises from candidate test questions according to teaching experiences. The method highly depends on the teaching and research experts and the professional level thereof, has overhigh cost and large fluctuation, and is not beneficial to large-scale popularization. In addition, in the interactive teaching process of the conventional AI interactive lesson, an evaluation link is usually set, weak knowledge points of students are diagnosed according to answer performances of the students in the evaluation link, and matched exercises of the weak knowledge points are recommended for the students to train, so that similar question groups with wrong answers are generally set in the evaluation link. The diagnosis method is too simple, the contingency of students in answering and evaluating questions cannot be avoided, the real weak points of the students are difficult to diagnose, the total answer amount of the students is increased due to the setting of the evaluating link, the efficient learning of the students is not facilitated, in addition, the recommendation link only recommends the similar question groups of the evaluating questions, the front-back dependency relationship among different knowledge points is often ignored, the difficulty progressive relationship among different exercise questions of the same knowledge point is ignored, the learning difficulty of the students is increased due to the lack of the progressive learning process, and the participation degree of the students is reduced.

Therefore, the invention provides a self-adaptive arrangement method of an AI interactive course. Fig. 9 is a flowchart illustrating a method for adaptively scheduling an AI interactive course according to the present invention, as shown in fig. 9,

s1, the lecture manually made by the instructor for the AI interactive lesson is used as small sample marking data, a knowledge system is built based on the corresponding relation between the knowledge points provided by the small sample marking data and a small number of matched test questions, and a test question knowledge point prediction model is trained based on the small sample marking data, so that the test question knowledge point prediction model can finally achieve the effect of being not lower than or close to the large sample marking data on the evaluation index. Based on the method, the massive resource question bank is used as the large sample unsupervised text topic data, the test question knowledge point prediction model is adopted to predict the knowledge points of the massive resource question bank, the massive resource question bank is associated with a knowledge system, test questions with low confidence coefficient are filtered out, and candidate test questions and knowledge points of the candidate test questions are obtained, so that the cost of manual labeling is greatly reduced.

And S2, extracting multi-modal course resources from the teaching video, particularly the stock teaching video generated in the online education scene, respectively extracting different modal characteristics of the audio, video and text data, and performing characteristic fusion on the extracted multi-modal characteristics through a model. Meanwhile, the same feature mapping is carried out on the candidate test questions with high confidence degree obtained in the knowledge point prediction link, and the feature extraction of the candidate test questions is completed. Mapping the video features and the test question features to the same semantic space, calculating the vector distance between the teaching video and each candidate test question in the space, intercepting the similar test questions corresponding to the teaching video when the vector distance is closer to indicate that the similarity between the teaching video and the candidate test questions is higher, and acquiring a knowledge point sequence formed by the knowledge points related to the teaching video based on the test question knowledge points of the similar test questions. And then organizing the learning sequence of the knowledge point sequence, namely the general learning path corresponding to the teaching video, according to the attribute information of the knowledge points, such as the exposure rate of the knowledge points, the difficulty level of the knowledge points, the teaching sequence of the knowledge points and the like in a large-scale examination, thereby realizing the organization of the general learning path corresponding to the teaching video according to the similarity relationship between videos and test questions, the mapping relationship between the test questions and the knowledge points and the association relationship between the knowledge points.

S3, a deep learning model, namely a user portrait model, is introduced to describe the implicit portrait of the user and is used for expressing the learning ability of the user, namely learning situation vectors of the user, wherein the closer the learning situation vectors are, the higher the similarity of the user is. And obtaining a learning situation vector of the target user based on the historical answer record of the AI interactive class student, namely the target user, and the user portrait model. Taking a knowledge point on the knowledge point sequence as an example, a user with historical answering behavior on the knowledge point is called a candidate user, searching a similar user group of a target user from the candidate user group according to the distance of the learning situation vector, and estimating the score of the target user on the knowledge point according to the average score of the similar user group on the knowledge point, thereby diagnosing whether the knowledge point is a weak knowledge point of the target user.

S4, generating an individualized learning path corresponding to the target user according to the weak knowledge point diagnosis result of the target user and the general learning path corresponding to the teaching video, recommending the learning path based on the individualized learning path, so as to guide the target user to selectively complete learning according to the learning path, and after the target user enters the individualized learning path, dynamically putting video slices, practice problem groups, test questions and the like according to the real answering condition of the target user at each knowledge point, so that invalid exercises are reduced and the learning efficiency is improved by dynamically updating the learning path.

The method provided by the embodiment of the invention can repeatedly utilize the teaching video which is generated by the online teaching scene and contains multi-mode data such as audio, video, text and the like, upgrade the common teaching of the live course scene into the individual teaching of the interactive course scene, reduce the education cost, promote the education fairness, relieve the burden of teachers and improve the efficiency of students.

The identification device of weak knowledge points provided by the present invention is described below, and the identification device of weak knowledge points described below and the identification method of weak knowledge points described above can be referred to in correspondence with each other.

Based on any one of the above embodiments, the embodiment of the present invention provides an apparatus for identifying a weak knowledge point. Fig. 10 is a schematic structural diagram of a weak knowledge point recognition apparatus provided in the present invention, and as shown in fig. 10, the apparatus includes:

the user portrait determining module 1010 is used for determining a user portrait of the target user based on the test question representation of each test question in the historical answer record of the target user and the historical score condition corresponding to each test question of the target user;

a similar user determination module 1020 for determining similar users of the target user based on the user representation of the target user;

a knowledge point determining module 1030, configured to determine weak knowledge points of the target user based on answer score conditions corresponding to the knowledge points of the similar users.

The device provided by the embodiment of the invention obtains the user portrait capable of representing the learning ability of the target user through the test question representation of each test question in the history answer record of the target user and the history scoring condition of the target user corresponding to each test question, determines the similar user of the target user based on the user portrait, and determines the weak knowledge point of the target user based on the history answer representation of the similar user corresponding to each knowledge point, thereby realizing the identification of the weak knowledge point of the target user without answering the target user at each knowledge point, avoiding the problem of increasing the answering workload of the user, avoiding the contingency of the user when testing and evaluating the questions, realizing the accurate identification of the weak knowledge point of the user, being beneficial to the targeted learning of the user, being also beneficial to the personalized learning resource recommendation of the weak knowledge point subsequently, the interpretability and acceptability of the learning resource recommendation are improved.

Based on any of the above embodiments, user representation determination module 1010 is configured to:

Based on any of the above embodiments, the knowledge points are determined based on the following steps:

Based on any one of the above embodiments, the test question knowledge points of the similar test questions are determined based on the following steps:

performing feature extraction on the similar test questions to obtain first text features of the similar test questions;

Based on any of the above embodiments, the video characteristics of the teaching video are determined based on the following steps:

The following describes the planning device for learned route provided by the present invention, and the planning device for learned route described below and the planning method for learned route described above can be referred to correspondingly.

Based on any one of the above embodiments, the embodiment of the present invention provides a planning device for a learned path. Fig. 11 is a schematic structural diagram of a device for planning a learned route according to the present invention, as shown in fig. 11, the device includes:

a user portrait determination module 1110, configured to determine a user portrait of a target user based on test question representations of the test questions in the historical answer record of the target user and historical scores corresponding to the test questions of the target user;

a similar user determination module 1120 for determining similar users of the target user based on the user representation of the target user;

a knowledge point determining module 1130, configured to determine weak knowledge points of the target user based on answer score conditions corresponding to the knowledge points of the similar users;

and the learned path planning module 1140 is configured to determine a learned path corresponding to the target user based on the weak knowledge points of the target user.

The device provided by the embodiment of the invention can display the test questions based on the test questions in the historical answer records of the target user, and the historical score condition of the target user corresponding to each test question, and determines the similar user of the target user based on the obtained user figure, and then determines the weak knowledge point of the target user based on the historical answer expression of the similar user corresponding to each knowledge point, thereby avoiding the problem of increasing the work load of answering questions of the user, avoiding the contingency of the user when answering and evaluating the questions, realizing the accurate identification of weak knowledge points of the user, on the basis, the learning path corresponding to the target user is determined based on the weak knowledge points of the target user, personalized learning resource recommendation can be performed aiming at the weak knowledge points, personalized teaching is achieved, the learning effect and efficiency of the target user are further improved, and meanwhile, the interpretability and acceptability of the learning resource recommendation are also improved.

Based on any of the above embodiments, the learned path planning module 1140 is configured to:

Fig. 12 illustrates a physical structure diagram of an electronic device, which may include, as shown in fig. 12: a processor (processor)1210, a communication Interface (Communications Interface)1220, a memory (memory)1230, and a communication bus 1240, wherein the processor 1210, the communication Interface 1220, and the memory 1230 communicate with each other via the communication bus 1240. Processor 1210 may invoke logic instructions in memory 1230 to perform a method of weak knowledge point identification, the method comprising: determining a user portrait of a target user based on the test question representation of each test question in the historical answer record of the target user and the historical score condition of the target user corresponding to each test question; determining similar users of the target user based on the user representation of the target user; determining weak knowledge points of the target user based on the answer score conditions of the similar users corresponding to the knowledge points;

or, to perform a method of learning path planning, the method comprising: determining a user portrait of a target user based on the test question representation of each test question in the historical answer record of the target user and the historical score condition of the target user corresponding to each test question; determining similar users of the target user based on the user representation of the target user; determining weak knowledge points of the target user based on the answer score conditions of the similar users corresponding to the knowledge points; and determining a learning path corresponding to the target user based on the weak knowledge points of the target user.

In addition, the logic instructions in the memory 1230 may be implemented in software functional units and stored in a computer readable storage medium when the logic instructions are sold or used as a stand-alone product. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.

In another aspect, the present invention further provides a computer program product, the computer program product comprising a computer program, the computer program being stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, a computer is capable of executing the method for identifying weak knowledge points provided by the above methods, the method comprising: determining a user portrait of a target user based on the test question representation of each test question in the historical answer record of the target user and the historical score condition of the target user corresponding to each test question; determining similar users of the target user based on the user representation of the target user; determining weak knowledge points of the target user based on the answer score conditions of the similar users corresponding to the knowledge points;

or, the method for planning the learned path provided by the above methods can be executed, and the method includes: determining a user portrait of a target user based on the test question representation of each test question in the historical answer record of the target user and the historical score condition of the target user corresponding to each test question; determining similar users of the target user based on the user representation of the target user; determining weak knowledge points of the target user based on the answer score conditions of the similar users corresponding to the knowledge points; and determining a learning path corresponding to the target user based on the weak knowledge points of the target user.

In yet another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored, the computer program, when being executed by a processor, implementing a method for identifying weak knowledge points provided by the above methods, the method comprising: determining a user portrait of a target user based on the test question representation of each test question in the historical answer record of the target user and the historical score condition of the target user corresponding to each test question; determining similar users of the target user based on the user representation of the target user; determining weak knowledge points of the target user based on the answer score conditions of the similar users corresponding to the knowledge points;

or, the method for planning the learned path by executing the above methods includes: determining a user portrait of a target user based on the test question representation of each test question in the historical answer record of the target user and the historical score condition of the target user corresponding to each test question; determining similar users of the target user based on the user representation of the target user; determining weak knowledge points of the target user based on the answer score conditions of the similar users corresponding to the knowledge points; and determining a learning path corresponding to the target user based on the weak knowledge points of the target user.

The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.

Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.

Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying weak knowledge points is characterized by comprising the following steps:

2. The method for identifying weak knowledge points according to claim 1, wherein the determining a user representation of a target user based on a test question representation of each test question in a historical test question record of the target user and a historical score corresponding to each test question of the target user comprises:

3. The method for identifying weak knowledge points according to claim 1, wherein the knowledge points are determined based on the following steps:

4. The method for identifying weak knowledge points according to claim 3, wherein the question knowledge points of similar questions are determined based on the following steps:

5. The method for identifying weak knowledge points according to claim 3, wherein the video characteristics of the teaching video are determined based on the following steps:

6. A method for planning a learned path, comprising:

7. The method for planning the learned path according to claim 6, wherein the determining the learned path corresponding to the target user based on the weak knowledge points of the target user includes:

8. An apparatus for identifying weak knowledge points, comprising:

9. A planning apparatus for learning a route, comprising:

10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the method for identifying weak knowledge points according to any one of claims 1 to 5 or the method for planning a learned path according to any one of claims 6 to 7 when executing the program.

11. A non-transitory computer-readable storage medium, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the steps of the method for identifying weak knowledge points according to any one of claims 1 to 5, or the method for planning a learned path according to any one of claims 6 to 7.