Search

CN-121998804-A - Knowledge-graph-based live broadcast auxiliary partner learning method and system

CN121998804ACN 121998804 ACN121998804 ACN 121998804ACN-121998804-A

Abstract

The application relates to the technical field of information, in particular to a live broadcast auxiliary partner learning method and system based on a knowledge graph. The method comprises the steps of collecting multi-mode learning behavior data of students in live broadcasting class, carrying out fusion modeling according to the multi-mode learning behavior data of histories to obtain a state recognition model, recognizing learning states of the students under current knowledge points according to the state recognition model, outputting understanding degree scores, locating the current knowledge points in a pre-built knowledge graph when the understanding degree scores are lower than a preset reference threshold, backtracking front key concept nodes according to knowledge dependency relations to obtain at least one knowledge path which hinders understanding of the knowledge points, and generating accompanying content according to the knowledge path, wherein the accompanying content comprises concise concept cards, dynamic demonstration videos, case comparison questions or knowledge migration exercises and pushing the accompanying content to student terminals.

Inventors

  • LIU TING

Assignees

  • 杭州思铺教育科技有限公司

Dates

Publication Date
20260508
Application Date
20260409

Claims (10)

  1. 1. The live broadcast auxiliary partner learning method based on the knowledge graph is characterized by comprising the following steps: Collecting multi-mode learning behavior data of students in live broadcasting class; performing fusion modeling according to the historical multi-mode learning behavior data to obtain a state identification model; Recognizing the learning state of the student under the current knowledge point according to the state recognition model, and outputting an understanding degree score; When the understanding degree score is lower than a preset reference threshold value, locating a current knowledge point in a pre-constructed knowledge graph, and backtracking a front key concept node according to a knowledge dependency relationship to obtain at least one knowledge path which hinders understanding of the knowledge point; And generating partner study contents according to the knowledge path, wherein the partner study contents comprise concise concept cards, dynamic demonstration videos, case pair questions or knowledge migration exercises, and pushing the partner study contents to a student terminal.
  2. 2. The knowledge-graph-based live broadcast assisted companion method of claim 1, wherein, The multi-modal learning behavior data includes: Student facial expression images or video streams collected through cameras, voice signals collected through microphones, classroom interaction event logs and screen operation track data, The student facial expression images or video streams are used for analyzing the attention and emotion states, the voice signals are used for extracting intonation, speech speed, pause and key word response characteristics, the classroom interaction event log comprises time stamps and contents of answering operations, hand lifting requests, chat comments, praise or questioning actions, and the screen operation track data comprise note recording actions, courseware page turning frequency, highlight mark areas and mouse moving paths.
  3. 3. The knowledge-graph-based live broadcast assisted companion method of claim 2, wherein, Fusion modeling is carried out according to the historical multi-mode learning behavior data, and the method for obtaining the state recognition model comprises the following steps: extracting the facial expression image or video stream by adopting a convolutional neural network to obtain visual attention and emotion characteristics; Extracting acoustic features of the voice signals to obtain a Mel spectrogram, and extracting intonation changes and language response modes by using a time sequence modeling network; carrying out event sequence coding on the classroom interaction event log to generate a structured behavior vector; Performing space-time clustering and behavior pattern mining on the screen operation track data, and extracting a learning input degree index; Taking the visual attention and emotion characteristics, intonation change and language response modes, structured behavior vectors and learning input indexes as input modes, and inputting the input modes into a newly built multi-mode fusion module, wherein the multi-mode fusion module dynamically weights contribution degrees of all the input modes by adopting a gating attention mechanism; Combining the marked knowledge point category labels and the understanding degree scores, and training the multi-mode fusion module from end to end, wherein the multi-mode fusion module outputs fusion characteristics with association relation with the understanding degree scores; and obtaining a state identification model according to the trained multi-mode fusion module and the association relation.
  4. 4. The knowledge-graph-based live broadcast assisted companion method of claim 3 wherein, The method for obtaining the association relation comprises the following steps: in the training stage of the multi-modal fusion module, aligning the fusion characteristics output by the multi-modal fusion module with the corresponding comprehension scores to construct a supervised learning target; calculating the deviation between the understanding degree score of fusion feature prediction and the marked understanding degree score by adopting a regression loss function or an ordered classification loss function; Optimizing parameters of the multi-mode fusion module through a back propagation algorithm, so that fusion features and understanding scores in an embedded space represent monotonic mapping or differentiable mapping relation; Introducing a contrast learning strategy, and applying distance constraint to sample pairs from the same knowledge points but with obvious difference in understanding degree scores; And solidifying a mapping function or weight parameter from the fusion feature to the understanding degree score in the trained multi-mode fusion module to serve as the association relation.
  5. 5. The knowledge-graph-based live broadcast assisted companion method of claim 3 wherein, The method for identifying the learning state of the student under the current knowledge point according to the state identification model and outputting the understanding degree score comprises the following steps: Checking whether the multi-modal learning behavior data acquired in real time is complete or not, if the multi-modal learning behavior data has a plurality of modal data loss, selecting a sample which is most matched with the acquired learning behavior data from a preset filling library for filling, wherein the preset filling library comprises a large amount of historical learning behavior data; inputting the filled multi-mode learning behavior data into a state recognition model, extracting key features reflecting the learning state of the student by using the response of the state recognition model, mapping the key features onto a predefined understanding degree scoring scale, and outputting the student understanding degree score for the current knowledge point.
  6. 6. The knowledge-graph-based live broadcast assisted companion method of claim 1, wherein, The method for constructing the knowledge graph comprises the following steps: receiving course standards, teaching material texts and teaching outlines, extracting subject knowledge point entities, and carrying out standardized naming on each knowledge point and associating unique identifiers; Identifying semantic dependency relations among knowledge points, wherein the semantic dependency relations comprise precedence relations, inclusion relations, causal relations, analogy relations and application scenes, and constructing directed edges among the knowledge points; forming a directed graph by taking the knowledge point entity as a node and the semantic dependency relationship as an edge, and taking the directed graph as an initial knowledge graph; Reading student history answer data and wrong question data, dynamically supplementing a hidden cognition dependent path through a graph embedding or association rule mining method, and quantifying side weights in an initial knowledge graph according to the mastering migration strength of students among knowledge points to obtain a final knowledge graph.
  7. 7. The knowledge-graph-based live broadcast assisted companion method of claim 1, wherein, The method for obtaining at least one knowledge path which hinders knowledge point understanding comprises the following steps of: Positioning a node corresponding to the current knowledge point in the knowledge graph as a target node; searching along the reverse direction of the knowledge dependent edge by taking the target node as a starting point, and extracting all reachable front knowledge point nodes to form a candidate dependent subgraph; Assigning a mastering probability value to each node in the candidate dependent subgraph by combining with the historical understanding degree scoring data of the students, wherein the mastering probability value is calculated based on the past answering accuracy and the review frequency of the students on the corresponding knowledge points; extracting a plurality of knowledge paths from the candidate dependent subgraphs, and calculating according to the edge weights, the node grasping probabilities and the path lengths to obtain grasping weaknesses of the knowledge paths; And generating at least one knowledge path which hinders understanding of the knowledge points according to a plurality of knowledge paths with the grasping weakness degree higher than a preset reference threshold value.
  8. 8. The knowledge-graph-based live broadcast assisted companion method of claim 1, wherein, The method for generating the partner study content comprises the following steps: extracting one or more nodes with the lowest mastering probability values on the path as an intervention focus; and matching and generating partner study contents from a pre-accessed teaching resource library according to the attribute of the intervention focus.
  9. 9. The knowledge-graph-based live broadcast assisted companion method of claim 8 wherein, Matching and generating partner study contents from a pre-accessed teaching resource library according to the attribute of the intervention focus, wherein the method for generating partner study contents according to the partner study contents comprises the following steps: When the attribute of the intervention focus is definition class or term class concept, generating a simple concept card with accompanying content, wherein the simple concept card comprises core definition, keyword interpretation and common misarea prompt; When the attribute of the intervention focus is procedural or principle knowledge, generating a dynamic demonstration video or a flow chart with accompanying learning content generated by calling a pre-accessed visualization engine, wherein the dynamic demonstration video or the flow chart displays knowledge evolution logic; The attribute of the intervention focus is a concept pair which is easy to be confused or needs to be resolved, a case comparison question with the accompanying study content as a structure is generated, and the case comparison question comprises a positive example, a negative example and a prompt for guiding students to conduct comparison analysis; the attribute of the intervention focus is a knowledge point pair with migration application value, and a variant problem or a cross-situation task with concomitant school content set in combination with the current teaching scene is generated.
  10. 10. The live broadcast auxiliary companion learning system based on the knowledge graph is characterized by comprising: the acquisition module acquires multi-mode learning behavior data of students in live broadcasting class; The modeling module performs fusion modeling according to the historical multi-mode learning behavior data to obtain a state identification model; The recognition module is used for recognizing the learning state of the student at the current knowledge point according to the state recognition model and outputting an understanding degree score; the backtracking module is used for positioning the current knowledge point in the pre-constructed knowledge graph and backtracking the front key concept node according to the knowledge dependency relationship when the understanding degree score is lower than a preset reference threshold value to obtain at least one knowledge path which hinders the understanding of the knowledge point; and the generation module is used for generating the partner study content according to the knowledge path, wherein the partner study content comprises a concise concept card, a dynamic demonstration video, a case pair question or knowledge migration exercise and is pushed to the student terminal.

Description

Knowledge-graph-based live broadcast auxiliary partner learning method and system Technical Field The application relates to the technical field of information, in particular to a live broadcast auxiliary partner learning method and system based on a knowledge graph. Background In the current background of rapid development of online education, live broadcasting class has become one of the mainstream teaching forms. However, the problem of unidirectional infusion and lack of personalized feedback commonly exists in traditional live broadcast teaching, and a teacher is difficult to sense the learning state of each student in real time and cannot accurately intervene in respect of individual cognitive impairment. Although with the development of artificial intelligence and online education in recent years, the existing live broadcast teaching system has a certain interactive function and personalized content generation. However, the technical problems still exist in teaching practice that 1) a real-time learning state recognition mechanism is lacked, 2) a traditional system cannot judge learning understanding conditions through student behavior data, so that a teacher is difficult to intervene in learning disorder in time, and meanwhile, inference support based on a knowledge structure is lacked, 3. The traditional AI auxiliary learning system is mostly based on keyword matching or question bank retrieval, cannot generate a personalized learning path according to a logical relation between knowledge, and cannot support a dynamic review and feedback mechanism. The teaching platform also often cannot automatically adjust the review rhythm according to the mastering degree of students, and the learning feedback is lagged and fragmented. In recent years, the development of multi-modal learning analysis (Multimodal LEARNING ANALYTICS, MLA) technology has provided new ideas for solving the above problems. By fusing multisource data such as facial expressions, voice intonation, interaction behaviors, operation tracks and the like, the attention, emotion and cognitive state of the student can be modeled more accurately. Meanwhile, the knowledge graph is used as an effective tool for structurally expressing a discipline knowledge system, so that the logical dependency relationship among knowledge points can be clearly described, and a foundation is laid for diagnosing the root cause of learning disorder and planning a personalized remedy path. Therefore, the live broadcast auxiliary learning accompanying system which can integrate multi-modal learning behavior recognition through multi-modal learning analysis and combine knowledge graph semantic reasoning with intelligent recommendation algorithm can be researched, so that real-time learning monitoring and targeted knowledge supplement can be realized in a classroom, and the teaching efficiency and learning experience can be improved. Disclosure of Invention Various embodiments of the present disclosure describe a knowledge-graph-based live broadcast assisted companion method and system. In a first aspect, an embodiment of the present disclosure provides a method for assisting in live broadcast accompanying learning based on a knowledge graph, including the steps of: Collecting multi-mode learning behavior data of students in live broadcasting class; performing fusion modeling according to the historical multi-mode learning behavior data to obtain a state identification model; Recognizing the learning state of the student under the current knowledge point according to the state recognition model, and outputting an understanding degree score; When the understanding degree score is lower than a preset reference threshold value, locating a current knowledge point in a pre-constructed knowledge graph, and backtracking a front key concept node according to a knowledge dependency relationship to obtain at least one knowledge path which hinders understanding of the knowledge point; And generating partner study contents according to the knowledge path, wherein the partner study contents comprise concise concept cards, dynamic demonstration videos, case pair questions or knowledge migration exercises, and pushing the partner study contents to a student terminal. In a second aspect, embodiments of the present disclosure provide a knowledge-graph-based live broadcast assisted companion learning system, including: the acquisition module acquires multi-mode learning behavior data of students in live broadcasting class; The modeling module performs fusion modeling according to the historical multi-mode learning behavior data to obtain a state identification model; The recognition module is used for recognizing the learning state of the student at the current knowledge point according to the state recognition model and outputting an understanding degree score; the backtracking module is used for positioning the current knowledge point in the pre-constructed knowledge graph and backt