Junior high school history auxiliary learning method and system based on knowledge graph

Abstract

The invention provides a junior high school history auxiliary learning method and system based on a knowledge graph, and belongs to the technical field of learning resource management and recommendation, and the method comprises the steps: obtaining junior high school history teaching data; the sources of the junior high school historical teaching data comprise a junior high school historical textbook, a test question book, an educational resource platform and a knowledge base; constructing a junior high school historical knowledge graph mode layer based on the junior high school historical teaching data; preprocessing the junior high school historical teaching data; constructing a junior high school historical knowledge graph data layer based on the preprocessed junior high school historical teaching data and the junior high school historical knowledge graph mode layer; and assisting students in learning based on the junior high school historical knowledge graph data layer. According to the invention, the informatization and intelligence level of history teaching in junior high schools can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of learning resource management and recommendation, and particularly relates to a junior high school history auxiliary learning method and system based on a knowledge graph. Background Art

[0002] At present, with the continuous in-depth development of educational informatization, junior high school history teaching faces many challenges and opportunities. Traditional junior high school history teaching mainly relies on textbooks, teacher explanations, and simple test questions. The knowledge presentation method is relatively single, and it is difficult for students to intuitively grasp the internal connections between historical knowledge, easily forming fragmented knowledge memories, resulting in insufficient understanding of the historical development context and complex relationships.

[0003] With the rise of artificial intelligence technology, as a semantic network technology, the knowledge graph can describe concepts, entities, and their relationships in the objective world in a structured form, and its application potential in the education field has gradually attracted attention. In the field of history, some research attempts to apply the knowledge graph to historical knowledge management and learning assistance. However, there are still many deficiencies in the existing technology. On the one hand, the construction of existing historical knowledge graphs often lacks a deep combination with the actual needs of junior high school history teaching and curriculum standards, making it difficult to accurately match the cognitive level and learning goals of junior high school students, resulting in limited practicality of the knowledge graph. For example, some knowledge graphs pursue the comprehensiveness and depth of knowledge too much, exceeding the understanding range of junior high school students and increasing the learning burden instead.

[0004] On the other hand, at the application level of the knowledge graph, most systems only stay at simple knowledge display or basic question-and-answer functions, failing to fully utilize the advantages of the knowledge graph in knowledge reasoning, relationship mining, etc., and unable to provide students with personalized and in-depth learning assistance. For example, existing historical question-and-answer systems often only start from a single knowledge point when answering questions, unable to combine multi-dimensional relationship information in the knowledge graph, resulting in answers lacking systematicness and relevance, and it is difficult to help students build a complete historical knowledge system.

[0005] In addition, the data sources of existing junior high school history learning assistance systems are relatively limited, and the data processing process is not fine enough, with problems such as knowledge redundancy, errors, or incompleteness, seriously affecting the quality of the knowledge graph and the learning assistance effect. For example, when collecting data, some systems do not strictly screen the content of educational resource platforms, and a large amount of low-quality and repetitive information enters the knowledge graph, interfering with students' learning.

[0006] In summary, how to construct a junior high school history auxiliary learning system based on a knowledge graph that closely conforms to the actual situation of junior high school history teaching, can efficiently integrate multi-source data, deeply mine historical knowledge relationships, and provide accurate and personalized learning assistance for students has become an urgent problem to be solved. Summary of the Invention

[0007] The objective of the present invention is to provide a junior high school history auxiliary learning method and system based on a knowledge graph.

[0008] The technical solution of the present invention is as follows: A junior high school history auxiliary learning method based on a knowledge graph, the method comprising: Obtaining junior high school history teaching data; the sources of the junior high school history teaching data include: junior high school history textbooks, test question books, educational resource platforms, and knowledge bases; Constructing a junior high school history knowledge graph schema layer based on the junior high school history teaching data; Preprocessing the junior high school history teaching data; Constructing a junior high school history knowledge graph data layer based on the preprocessed junior high school history teaching data and the junior high school history knowledge graph schema layer; Assisting students in learning based on the junior high school history knowledge graph data layer.

[0009] Further, the constructing of the junior high school history knowledge graph schema layer based on the junior high school history teaching data is specifically as follows: Sorting out the historical timeline based on the junior high school history teaching data, abstracting historical figures, historical events, geographical concepts, and political and economic system concepts, and classifying them into ancient, modern, and contemporary; Inviting front-line junior high school history teachers, and adjusting and summarizing junior high school history concepts including 11 concept categories in combination with curriculum standard interpretation, teaching practice, and the five core historical qualities; Defining object attributes and data attributes for the junior high school history concepts, and constructing a junior high school history knowledge graph schema layer.

[0010] Further, the 11 concept categories of junior high school history concepts specifically include: Historical period, historical figure, historical event, historical region, historical group, politics, economy, civilization creation, historical document, art work, ideological theory.

[0011] Further, the junior high school history knowledge graph schema layer specifically includes: An object property system, used to express the relationships between different entities, including: there is a "successor" relationship between historical periods; there is a "being in" relationship between historical figures and historical periods; there is a "kinship, related to" relationship between historical figures and historical figures; there is a "participation, involvement" relationship between historical figures and historical events; there is a "belonging to" relationship between historical figures and historical groups; there is a "being in" relationship between historical events and historical periods; there is a "taking place in" relationship between historical events and historical regions; there is a "containing" relationship between historical events and politics, economy; there is a "containing, related to" relationship between historical events and historical events; there is a "participating" relationship between historical groups and historical events; there is a "existing in" relationship between historical groups and historical periods; there is a "being located in" relationship between historical groups and historical regions; there is a "creator" relationship between civilization creations and historical figures; there is a "development" relationship between civilization creations and civilization creations; there is a "political association" relationship between politics and politics; there is a "correlation" relationship between politics and economy; there is a "related personnel" relationship between politics and historical figures; there is a "related organization" relationship between politics and historical groups; there is an "economic association" relationship between economy and economy; there is a "related personnel" relationship between economy and historical figures; there is a "related organization" relationship between economy and historical groups; there is an "introduction" relationship between historical documents and civilization creations; there is an "author" relationship between historical documents and historical figures; there is a "reflection" relationship between historical documents and ideological theories; there is a "creator" relationship between art works and historical figures; there is a "creative prototype" relationship between art works and historical events; there is a "proposer" relationship between ideological theories and historical figures; there is a "proposing group" relationship between ideological theories and historical groups; The data attribute system is used to describe various concepts in detail. Among them, the data attributes of historical periods include period overview, specific years, political system, capital city, territory, status, learning objectives, and resource links; the data attributes of historical figures include alias / title, date of birth, date of death, achievements, positions, works, learning objectives, and resource links; the data attributes of historical events include background, time, process, results, impacts, significance, learning objectives, and resource links; the data attributes of historical regions include location, description, learning objectives, and resource links; the data attributes of historical groups include existence period, description, location, representative figures, status, learning objectives, and resource links; the data attributes of politics include period, creator / creating organization, concept, manifestation, role, learning objectives, and resource links; the data attributes of economy include period, proposer / proposing organization, concept, manifestation, role, learning objectives, and resource links; the data attributes of civilization creation include construction time / invention time, constructor / inventor, description, significance, learning objectives, and resource links; the data attributes of historical documents include author, time, brief introduction, content, significance, learning objectives, and resource links; the data attributes of art works include creator, time, brief introduction, significance, learning objectives, and resource links; the data attributes of ideological theories include time of proposal, proposer, generalization, significance, evaluation, learning objectives, and resource links.

[0012] Furthermore, the preprocessing of the junior high school history teaching data is specifically as follows: For junior high school history teaching data, highly homogeneous and low-quality content is removed in combination with expert opinions; then it is converted into text information. The text information is normalized, punctuation marks that contribute nothing to semantic understanding are removed, special characters, blank lines, and line breaks are cleaned. After cleaning, the text is segmented, and the generated sentence list is converted into a format with one sentence per line to generate a TXT text.

[0013] Furthermore, constructing the data layer of the junior high school history knowledge graph based on the preprocessed junior high school history teaching data and the pattern layer of the junior high school history knowledge graph is specifically as follows: Entity extraction, relationship extraction, and attribute extraction are performed on the preprocessed text data to convert junior high school history teaching data into structured knowledge. Among them, entity extraction is based on the model labeled in the pattern layer, relationship extraction adopts a combination of model and manual methods, and attribute extraction selects encyclopedia acquisition, large model extraction, or manual addition according to different attribute characteristics. A method combining attribute similarity and structure similarity is used for knowledge fusion. Potential duplicate entities are screened out by calculating similarity, and after manual verification, redundancy and conflicts are eliminated. Store the fused knowledge in the Neo4j graph database, represent historical knowledge entities as nodes, and represent the relationships between entities as edges to construct the data layer of the junior high school history knowledge graph.

[0014] Furthermore, assisting students' learning based on the data layer of the junior high school history knowledge graph is specifically as follows: For the questions input by students, first conduct domain determination to judge whether it belongs to the category of the history subject; if it belongs, extract the question keywords, retrieve relevant information in the knowledge graph, and optimize the obtained information after retrieval to remove redundant content; integrate the optimized information with the original question, and if there is a dialogue history, refine the key information of the dialogue history and combine it with the original question to form a comprehensive prompt template and input it into the large model to generate an answer.

[0015] The present invention also provides a junior high school history assisted learning system based on a knowledge graph, and the system includes: Data acquisition module: responsible for acquiring junior high school history teaching data from junior high school history textbooks, test question books, educational resource platforms, and knowledge bases; Mode layer construction module: Based on the acquired junior high school history teaching data, sort out the historical timeline, abstract and classify historical concepts, invite teachers to adjust and summarize the concept system, define object attributes and data attributes, and construct the mode layer of the junior high school history knowledge graph.

[0016] Data preprocessing module: Process the acquired junior high school history teaching data, remove highly homogenized and low-quality content, convert various formats of data into text information, and then normalize the text information to generate TXT text for subsequent use; Data layer construction module: Based on the preprocessed data and the mode layer, perform knowledge extraction, knowledge fusion, and knowledge storage to construct the data layer of the junior high school history knowledge graph; Learning assistance module: Based on the data layer of the junior high school history knowledge graph, implement the knowledge Q&A function Furthermore, the learning assistance module includes: User management module: Support students and teachers to log in; students are used to complete information and view notes; teachers are used to manage students' accounts, view learning progress, edit the knowledge graph and knowledge base; Knowledge graph display module: Visually present the knowledge graph, and users can zoom in, drag, and browse the whole picture, and filter specific types of knowledge points through legends to view the knowledge context; Knowledge graph query module: Students conduct sub-graph and knowledge point queries.

[0017] Knowledge graph editing module: Teachers or administrators perform addition, deletion, and modification operations on the knowledge points and relationships of the knowledge graph; Knowledge base module: Store learning resources by grade. Students can select their grade to study, view resources, related knowledge points, and take notes; teachers or administrators can upload resources and edit knowledge points. Historical Q&A module: Combine the knowledge graph and large model to determine the question domain, retrieve information, and generate answers.

[0018] Compared with the prior art, the present invention has the following advantages: In terms of knowledge construction, it closely follows the People's Education Edition textbooks and curriculum standards. Through the participation of front-line teachers, a knowledge graph is constructed in combination with historical core qualities to ensure the accuracy of knowledge and teaching adaptability. In terms of data processing, multi-source data is collected and finely preprocessed to ensure data quality. In terms of functional application, knowledge Q&A combines large models and knowledge graphs to provide accurate answers; multi-module functions support students' personalized learning and teachers' efficient management. For example, the knowledge graph display and query module helps students intuitively understand the knowledge context, and the knowledge base module facilitates resource management and learning. Compared with traditional methods, the present invention presents historical knowledge in a structured and visual way, enhancing the learning effect and improving the informatization and intelligence level of junior high school history teaching. Brief Description of the Drawings

[0019] The drawings generally illustrate various embodiments by way of example and not limitation, and are used together with the description and claims to explain the embodiments of the invention. Where appropriate, the same reference numerals are used throughout the drawings to refer to the same or similar parts. Such embodiments are illustrative and are not intended to be an exhaustive or exclusive embodiment of the device or method.

[0020] Figure 1 Shows the framework diagram of the construction of the junior high school history knowledge graph of the present invention; Figure 2 Shows the ontology visualization diagram of the present invention; Figure 3 Shows the knowledge extraction flow chart of the present invention; Figure 4 Shows the schematic diagram of the implementation process of the Q&A function of the present invention; Figure 5 Shows the schematic diagram of the system function of the present invention. Detailed Description of the Embodiments

[0021] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. The following will refer to the drawings and combine the embodiments to detail the present application.

[0022] The embodiments of the present invention provide a junior high school history auxiliary learning method and system based on a knowledge graph. The method includes: 1 Framework for constructing the junior high school history knowledge graph and data sources 1.1 Construction Framework

[0023] The construction framework of the junior high school history knowledge graph mainly consists of three parts: defining the schema layer, constructing the data layer, and storing the knowledge graph, as specifically shown Figure 1 below.

[0024] Defining the schema layer: The entity types and relationships involved in junior high school history knowledge are complex and diverse, including historical periods, historical figures, historical events, historical regions, etc. When defining the schema layer, it is necessary to analyze and determine these entity types and their attributes in detail, and clarify their diverse relationships to ensure that junior high school history knowledge can be described comprehensively and accurately. The schema layer is the foundation of the junior high school history knowledge graph. At this stage, a set of description specifications need to be formed to serve as the core of graph construction.

[0025] Constructing the data layer: The construction of the data layer is based on the schema layer and mainly consists of two stages: knowledge extraction and knowledge fusion. The knowledge extraction part extracts entities, relationships, and their attributes from text data according to the standards set by the schema layer, providing data support for graph construction. In the knowledge fusion part, there is a large amount of redundancy and repetition in the extracted content, and it is necessary to eliminate semantic ambiguous entity words in the corpus to ensure that the entire knowledge graph has a unified identifier and knowledge association.

[0026] Storage of the knowledge graph: Some knowledge of junior high school history also develops continuously over time, and it is necessary to update the knowledge graph regularly to reflect the latest history textbooks and literature research. The storage and update mechanism of the knowledge graph needs to be flexible and efficient, capable of quickly integrating new knowledge, and storing junior high school history knowledge information in the Neo4j graph database for updating, querying, and visual display.

[0027] 1.2 Data Sources Constructing a junior high school history knowledge graph must rely on authoritative, rigorous, and standardized data support. This study mainly uses resources such as junior high school history textbooks, test question books, and educational platforms to guide data screening and construct a conceptual framework to ensure the unity of knowledge representation in the knowledge graph. According to the media differences of junior high school history subject resources, they can be divided into entity resources and digital resources.

[0028] Entity resources refer to traditional printed materials that exist in physical form, specifically including the history textbooks for grades seven, eight, and nine of the People's Education Edition, the history teacher's teaching reference books for grades seven, eight, and nine of the People's Education Edition, the "Compulsory Education Curriculum Plan and Curriculum Standards (2022 Edition)", the real exam questions of history for the high school entrance examination in the past five years, stage practice test papers, and history workbooks published by Zhejiang Education Publishing House, etc.

[0029] Digital resources are resources stored and disseminated in digital form, which can be divided into two categories: educational resource platforms and knowledge bases. Specifically, they include educational resource platforms such as 21st Century Education Network, Middle School History Teaching Garden, and National Smart Education Platform for Primary and Secondary Schools; knowledge bases such as Baidu Encyclopedia and the basic education knowledge service platform developed by the Knowledge Engineering Laboratory of Tsinghua University.

[0030] 2 Construction of the Pattern Layer of Junior High School History Knowledge Graph The basic component unit of a knowledge graph is the triple of "entity - relationship - entity" and "entity - attribute - value", which is a knowledge network that connects entities, attributes, and their relationships. The core of the pattern layer of the junior high school history knowledge graph is to construct a knowledge concept ontology. The ontology model clearly defines and formally describes the entity types of junior high school history knowledge and the relationships between entities. Through this ontology model, the rules for knowledge extraction in the data layer are restricted and standardized, reflecting the knowledge framework of junior high school history. This section mainly includes determining entity types and attributes, defining relationships between entities, and ontology visualization based on Protégé, providing a standardized data model for knowledge extraction in the data layer.

[0031] 2.1 Determining Entity Types First, based on the six textbooks of junior high school history and teacher's teaching reference books of the People's Education Edition, the historical timeline is sorted out, and then concepts such as historical figures, historical events, geographical concepts, political and economic systems, etc. are abstracted around this timeline. To improve concept extraction, it is classified according to historical periods of ancient, modern, and contemporary to ensure that the important concepts and terms summarized can be applied to each period. Then, front-line junior high school history teachers are invited to adjust and summarize a set of junior high school history concept and term systems in combination with the interpretation of the History and Society curriculum standards and actual teaching practices, and based on the five core historical qualities: materialist view of history, historical evidence, sense of time and space, historical interpretation, and patriotism and family feelings. It contains a total of 11 concept categories. To better describe the composition of the core categories, the four core categories of historical groups, politics, economy, and civilization creation summarized are further subdivided.

[0032] Table 2.1 Ontology Concept Categories , 2.2 Defining Entity Relationship Types and Attributes Attributes in the ontology refer to the relationships used to describe the characteristics and behaviors of classes, which can be divided into two categories: object properties and data properties. Object properties are used to express the relationships between different entities, such as "participate in", "be in", and "be located in", etc. Starting from the defined core classes in this section, combined with expert opinions, the relationships of core classes in historical resources are analyzed, and a systematic object property system is constructed following consistency and simplicity, as shown in Table 2.2 below.

[0033] Table 2.2 Object Property Classification , Data attributes refer to the detailed description of a class, and the ontology construction is improved by defining multiple attribute tags. In the field of junior high school history, introducing data attributes can not only provide students with rich learning materials, but also help to construct a more comprehensive historical knowledge learning framework. For example, in the category of historical events, data attributes such as background, time, process, result, influence, significance, relevant materials, learning objectives, and resource links are added to jointly form a three-dimensional historical event model. According to the characteristics of different categories, this study defines matching data attributes for each category, as shown in Table 2.3 below.

[0034] Table 2.3 Classification of Data Attributes ,

[0035] Finally, the protégé software is used in the research to realize the knowledge modeling of the junior high school history ontology, as shown specifically in Figure 2 . Protégé is a free and open-source ontology editing tool developed by Stanford University, which supports multiple ontology description languages such as OWL, RDF, and RDFS, and has functions such as class modeling, instance editing, and model processing. The graphical interface it displays is user-friendly.

[0036] 3 Construction of the Data Layer of the Junior High School History Knowledge Graph 3.1 Source Data Preprocessing The acquisition channels of educational resources are becoming increasingly diversified. In this section, OCR software is used to batch recognize physical resources such as textbooks and teacher's teaching reference books, and convert them into editable TXT text format. Some of the resources such as courseware and videos used to supplement the dataset come from the support of front-line teachers, and the other part is screened and downloaded from educational websites such as the Middle School History Teaching Garden. Since there is a shortage of up-to-date free resources on some websites, manual screening and downloading are carried out. For platforms with relatively rich free resources, a crawler algorithm is selected to obtain them in batches. During the data source processing, first, the five types of files, namely videos, PPTs, WORDs, PDFs, and TXTs, obtained are manually screened, and highly homogeneous and low-quality content is removed in combination with expert opinions. The video content is converted with the help of the artificial intelligence application Tongyi Tingwu to comprehensively save the video information. Subsequently, the ExtractPDF tool is used to separate and extract text and pictures from PPT and PDF files, and finally, TXT files containing text information are obtained to realize the classification processing of different types of data sources. Finally, regular expressions and string methods are used to normalize the text, remove punctuation marks that contribute nothing to semantic understanding, clean special characters, blank lines, and line breaks, etc. After cleaning, the text is segmented, and the generated sentence list is converted into a format with one sentence per line and stored in a TXT file.

[0037] 3.2 Junior High School History Knowledge Extraction and Fusion Knowledge extraction is the core part of constructing the junior high school history knowledge graph. It involves obtaining key knowledge from diverse resources and transforming this knowledge into a structured form for storage in the knowledge graph. This process encompasses three core tasks: entity recognition, relation extraction, and attribute extraction. The process is as follows Figure 3 as shown

[0038] This study takes junior high school history as the research object. Considering the wide applicability of historical discipline knowledge and the relatively small scale of the corpus in this study, constructing a new knowledge extraction model is not a necessary step. In view of this, this study uses the Baidu PaddleEasyDL platform to implement knowledge extraction. The EasyDL platform is an AI development tool based on the open-source deep learning platform Baidu PaddlePaddle, providing an integrated artificial intelligence model training and deployment environment for academic research and industrial practice, covering the entire cycle of AI model development, including data preparation, model training, and deployment applications. Among them, the platform integrates the ERNIE 3.0 version of the Wenxin large model for text processing and analysis scenarios, with an average accuracy of over 90% in various NLP tasks. In addition, the platform uses transfer learning technology and AutoDL technology to achieve rapid fine-tuning and optimization of the model to meet different data characteristics and task requirements.

[0039] (1) Entity extraction This study uses some preprocessed and cleaned TXT texts to form the junior high school history entity recognition dataset. The dataset contains historical data from three grade levels, covering domestic and foreign, ancient, modern, and contemporary content. The data annotation process is divided into two parts: creating a label group and online annotation. Combining the previous classification research on the junior high school history ontology, by creating a label group, these categories and subcategories are incorporated into the annotation system as entity labels. Each label corresponds to a specific entity type, providing clear guidance for entity extraction. The model training preferably selects high-precision algorithms, and further improves the prediction performance of the model by integrating the Wenxin large model, and can also perform well with a small number of samples. After the model training is completed, the platform conducts an overall evaluation of it. The results show that the F1-Score value is 89.9%, the precision is 88%, and the recall is 92%, indicating that the model performs excellently in the entity extraction task and achieves high accuracy and coverage.

[0040] The public cloud API deployment method better meets the research requirements with its efficient remote call function, perfect security mechanism, and the feature of not having to worry about hardware configuration. Therefore, the public cloud deployment method is selected in the research to implement model calls. After authentication debugging, the input TXT text is processed twice in Python to make it meet the model input requirements. At the same time, according to the field types, for example, the field of a specific entity is span, and the field of the corresponding entity category is tag, the field output format is adjusted and written to a file; After the model extraction is completed, in order to improve the accuracy and comprehensiveness of the obtained entities, based on junior high school history textbooks and combined with the professional suggestions of junior high school history teachers, the misclassified entity types are corrected, the completely duplicate entities are removed, and the missing and necessary entities are supplemented.

[0041] (2) Relationship extraction The method of relationship extraction is similar to the entity extraction method described above, and both use the EasyDL platform for operation. When performing entity relationship extraction on the platform, a dataset mainly composed of comprehensive documents and teacher's teaching reference books is used. Compared with textbooks, they have a more structured language expression and clearer relationships between entities. In this study, the relationship types are divided into two categories according to the complexity of the relationship and the dependence on context. One category is suitable for extraction through an automated platform, and the other category requires manual extraction based on expert prior knowledge. The former usually has clear clues and often has a relatively fixed expression in the text, which is convenient for the algorithm to identify. For example, the "occurred in" relationship between "historical event" and "historical region", and the "exists in" relationship between "historical group" and "historical period", etc. The latter involves relatively complex logic and requires in-depth semantic understanding to complete the identification. On the one hand, if there is no clear expression such as "wooden movable type is the development of clay movable type" in the text, the probability of correctly identifying the relationship will decrease. For example, the "associated with" relationship between "politics" and "economy", and the "development" relationship between "civilization creation" and "civilization creation", etc. On the other hand, in the "related to" relationship between "historical event" and "historical event", "related" is a relationship that depends on the specific context and also requires expert judgment and interpretation. Therefore, the relationship extraction adopts a method combining model and manual work. The platform processes the relationships that are relatively fixed and easy to identify, and manual extraction supplements the relatively complex relationships that require in-depth understanding to ensure that the quality of the constructed knowledge graph meets the application requirements of junior high school history.

[0042] Among them, the model extraction part is evaluated by three indicators: precision, recall, and harmonic mean. The F1-Score value is 85.5%, the precision is 82%, and the recall is 89.4%. The results show that the model has achieved good results in the relation extraction task. Subsequently, it is released to the public cloud service. In the Python environment, the API is called and the output parameters are set, including fields such as the entity content span, entity category tag, and entity relation type predicate. Finally, the output content is written to a file.

[0043] To ensure the accuracy and applicability of the relations, this study invited junior high school history teachers to examine and correct the results obtained from the model and manual extraction. At the same time, the correctness, educational significance, and clarity of the expressions of the relations were evaluated and optimized. Through this process, a solid foundation was provided for constructing the junior high school history knowledge graph.

[0044] (3) Attribute extraction Attribute extraction refers to the process of extracting entity attributes from text data, which involves the identification and extraction of descriptive information of entities. Complete attribute values help to comprehensively understand entity characteristics. In this study, three methods are used for extraction according to different attribute characteristics, namely obtaining from encyclopedia data, extracting using a large model based on a knowledge base, and manual extraction.

[0045] Obtaining from encyclopedia data: Taking the historical figure entity class as an example, the attributes of "alias / title", "date of birth", "date of death", and "works" in Baidu Encyclopedia correspond to the "alias", "date of birth", "date of death", and "major works" of the entry respectively. The attribute data can be obtained by writing a crawler algorithm. By integrating the title, overview, and information box modules in the entry, the title provides an intuitive description of the entry, the overview covers the core information of the entry, and the information box shows specific attributes in a structured form. It is written to an Excel file in the format of one title and the corresponding overview and information box per line and stored locally.

[0046] Knowledge base extraction: Taking historical event entities as an example, for the "background", "process", "result", "influence", and "significance" among them, one or more sentences are required to describe them, and it is difficult to obtain attribute values through traditional annotation methods. However, in-depth analysis can be carried out based on documents with clear prompt words to improve the effect of attribute extraction. Based on the special review materials, teacher's teaching reference books, test questions and their answers obtained in the data collection stage, an unstructured knowledge base is formed. Compared with textbook materials, these types of data have more advantages in the entity attribute organization method. Secondly, use the entity overview part in the crawled encyclopedia data and store it in tabular form according to the format of entity name and overview to form a structured knowledge base, ensuring that attributes that cannot be obtained in the unstructured knowledge base can be supplemented from other sources. Finally, use Zhipu Qingyan to combine the unstructured and structured knowledge bases to achieve attribute extraction. And output the extraction results in a structured tabular form, and save it as a tabular form using the text import wizard in Excel by delimiter.

[0047] Manual extraction: The integration of learning objectives and resource links is also included in entity attributes. The learning objectives are set according to Bloom's classification of the cognitive domain levels, including six cognitive levels: remembering, understanding, applying, analyzing, synthesizing, and evaluating, following the progression of cognitive levels. The research mainly improves the learning objective attributes based on the "Compulsory Education History Curriculum Standard (2022 Edition)" published by Beijing Normal University Press. The resource link attributes are added based on the links and their titles obtained during the resource collection process. For some resources corresponding to specific class hours in the textbook, compare them with the unit content in the textbook and add these resource links to the resource link attributes of the corresponding historical period entities. Through the mapping and association between entities and resources, a junior high school history learning resource network that meets teaching needs and is convenient for students to explore is constructed.

[0048] (4) Knowledge integration Knowledge integration refers to the unification and standardization of information in the knowledge base by integrating the ontology layer and the data layer. In this study, only the data layer is used for knowledge integration, that is, the entity alignment process. Among a large number of historical entities, the high repetition of entity names can be processed through simple screening. However, for those entities that refer to the same thing but use different expressions, such as "Sima Rui" and "Emperor Yuan of Jin" in historical figures, "Aisin Gioro Yixin" and "Prince Gong" etc., it is necessary to design an entity similarity evaluation method to integrate these implicitly repeated entities. In this study, a calculation formula is designed by comprehensively considering attribute similarity and structural similarity to evaluate the combined similarity between two entity records A and B.

[0049] , Attribute similarity The calculation of adopts the Jaccard similarity to measure. First, it is necessary to determine the neighbor set of the entity, that is, all entities directly connected to this entity. In order to effectively screen entities, a chunk comparison strategy is adopted. By dividing the entities into several subsets or chunks according to the textbook content order set first, the structural similarity is compared within each chunk, and the similarity of entity pairs is compared through a double loop and stored in a matrix. Finally, according to the calculation formula of the comprehensive attribute similarity and the structural similarity, potential duplicate entities with a similarity exceeding 0.7 are screened out. Finally, it is judged whether these entities are the same entity through manual verification, and which entity to retain in the knowledge graph is determined.

[0050] 3.3 Knowledge Storage Considering the importance of entity relationships in the junior high school history knowledge graph, this study uses the Neo4j graph database for storage. Neo4j can not only clearly represent the multi-hop connections between entities and their attributes, but also quickly retrieve, screen, and modify the graph content through statements. With the help of the Py2neo library to interact with the Neo4j graph database to achieve the import of structured data, finally 4,180 nodes and 6,648 relationships are obtained.

[0051] 4 Core Function Design of the Learning Assistant System After the construction of the knowledge graph is completed, how to apply it intelligently is the core goal of the function research of the learning assistant system. The retrieval, editing, and visualization functions based on the graph and the derived function modules can be implemented in the subsequent front-end and back-end interactions of the system, specifically through the graph database operation library and the corresponding query language. However, the knowledge Q&A function, as the core module of the learning assistant system, needs to be studied and experimented in advance in this section, because the implementation of knowledge Q&A not only depends on the structured information of the knowledge graph, but also needs to combine appropriate natural language processing technologies to accurately understand the user's query and generate answers.

[0052] Traditional question-answering implementation based on knowledge graphs usually relies on intent recognition and entity recognition to build rules and templates for retrieving answers from knowledge graphs. However, this preset-rule-based method has obvious deficiencies in the richness and flexibility of answering questions. When the system faces complex, open-ended, or logically reasoning questions, due to the lack of sufficient semantic understanding and reasoning capabilities, the generated answers are often single and lack depth, limiting the application of question-answering in educational scenarios and the improvement of user experience. Given the excellent performance of large models in question-answering tasks in recent years, this study combines the advantages of junior high school history knowledge graphs and large models to achieve the question-answering function of accurately answering basic knowledge questions related to junior high school history courses and handling some historical interpretation questions that require comprehensive analysis and reasoning. Specifically, the RAG method is used to combine the knowledge graph and the large model to implement the junior high school history question-answering function. To build a complete link and achieve the deep integration and application of the knowledge graph and the large model, this study also introduces the engineering framework - LangChain, which can combine large language models with external diverse data sources, thereby broadening the application scenarios of the model. The flowchart of the question-answering function implementation is specifically as Figure 4 shown.

[0053] 4.1 Knowledge Question-Answering Implementation (1) Environment Preparation Before starting to build the question-answering function, it is also necessary to clarify the selection of the large model. This study compared large models in the Chinese field and selected GLM-4-Plus (Zhipu Qingyan series), ERNIE4.0 Turbo (ERNIE large model series), and Qwen-Max (Tongyi Qianwen series) to implement question-answering respectively. GLM-4-Plus is a large language model launched by Zhipu AI, adopting a sparse MoE architecture with a dynamic gating mechanism. Compared with the fully connected structure of its predecessor GLM-3, its reasoning and instruction-following capabilities have been significantly improved. ERNIE 4.0 Turbo is a large language model developed by Baidu Research Institute and is the latest version of the ERNIE series. It uses a vocabulary of 8K context length in the pre-training stage to enhance the model's ability to handle complex language phenomena. Qwen-Max is a large language model launched by Alibaba Cloud, adopting a hierarchical sparse attention (HSA) mechanism and a language adaptive embedding (LAE) technology, which enhances the semantic coherence between paragraphs and the model's ability to understand and respond to human intentions in long text understanding tasks. At the same time, it is also necessary to obtain the API access rights of the selected large model, which involves applying for API keys, setting API call methods, etc.

[0054] (2) Determine the Domain The Q&A function developed by this research institute aims at the knowledge field of junior high school history. Therefore, for the input questions, it is necessary to perform question classification operations first to determine whether the questions belong to the category of history discipline, so as to ensure that the system response conforms to the target discipline field and effectively improve the accuracy and relevance of Q&A. Define the prompt string for the large language model, and the task directly provided to it is to judge whether the user's question is related to the history discipline. Further refine the task scope. According to the content, keywords and context of the question, as well as the provided examples of history-related keywords, it is pointed out that the question must be related to ancient Chinese history, modern Chinese history, ancient world history, and modern world history. Then, return a JSON-formatted answer according to the specific template requirements. Then, use the system prompt and the user's question to call the corresponding language model, and parse the JSON response from the content. If the result shows that it is related to the history discipline, pass the user's question as a parameter to the structured_retriever function for subsequent keyword extraction and retrieval. If the result is judged to be irrelevant to the history field, prompt the user to reorganize the language and ask questions related to history for Q&A.

[0055] (3)Keyword Extraction Extract keywords for questions that meet the domain conditions. By extracting keywords, corresponding entities can be retrieved more accurately in the knowledge graph. In this research, according to the classification in entity recognition, an example part of the prompt template for keyword extraction is designed:

Template

[0056] Reply format: Entity 1, Entity 2, …… Examples of keywords for reference: (historical figures): Qin Shi Huang, Confucius, Napoleon; (historical events): the 1911 Revolution, the Renaissance, the Pearl Harbor Incident… (organizations): the United Nations, the Holy See, the Communist International… (political situations): the Kangxi-Qianlong Prosperity, the bipolar pattern, the tripartite confrontation of the Three Kingdoms… (art works): Mona Lisa, Symphony No. 9 (ideological theories): Marxism, universal love and non-attack.

[0057] However, if no valid information is retrieved from the knowledge graph, based on the preliminary answer text generated by the large model for the original question without prompts, the key entity extraction template described above is used to perform key entity extraction on the answer text. Through re-information extraction, the identified key entities are more in line with the context of the potential answer compared to the entities in the initial question. Also, when these entities are retrieved in the knowledge graph, it is more likely to find content closely related to the question. Although extracting key entities from the preliminary answer can improve the accuracy of the retrieval, this process increases the system's response time. Therefore, the branching method is only used when the question itself does not provide valid entities.

[0058] (4)Retrieve the knowledge graph Through the keyword extraction step, key entities are obtained from the question text or the preliminary answer of the large model. Next, the entities need to be input into the knowledge graph for matching. Implementing this process mainly includes four steps: database connection, relationship query, node information extraction, and data formatting and storage. First, initialize the Graph object to establish a connection with the Neo4j graph database, specifying the Bolt protocol endpoint and authentication credentials of the database. Then define a function to receive a text parameter containing the key entities. Inside the function, define two Cypher query strings, one for matching the outgoing relationships starting from the node identified by the specified name and the other for matching the incoming relationships pointing to the node. During the query execution phase, first execute the Cypher query for the outgoing relationships to retrieve the relationships from the specified node to other nodes. The retrieval results are stored in a variable, and a loop structure is used to iterate through each query result. For each iterated element, extract the relationship entity and construct a formatted string that includes the label set, property value set, relationship type name of the starting node, as well as the label set and property value set of the ending node. Similarly, execute the query for the incoming relationships and repeat the above process to generate a formatted string.

[0059] When the information retrieved from the knowledge graph is input into the large model, this information, as background knowledge and prompts, can improve the fit of the model with the actual teaching content when answering questions. However, there may be a certain degree of redundancy in the content and its relationships extracted from the knowledge graph. Especially when the number of key entities is large, or when the entities retrieved from the graph and their associated entities are rich, the amount of data input into the large model may increase significantly. In view of this, it is necessary to further optimize the input content after the knowledge graph retrieval process. Input the retrieved information and the original question together, and filter and streamline them through a predefined template to remove redundant information and retain key content, ensuring that the data finally input into the large model is sufficient and efficient to guarantee the accuracy and context coherence of the model's generated answer.

[0060] (5)Integrate the context If a conversation history exists during a continuous dialogue interaction, a parallelization strategy is employed to process the conversation history and question text. Key information from the conversation history is extracted through a conversation compression process. A prompt framework is then used to combine this extracted information with the original question to form a new question independent of the conversation history. If no conversation history exists, the original question is used directly. Finally, the information extracted from the knowledge graph is combined with the question to form a comprehensive prompt template.

[0061] 4.2 Question-Answering Effectiveness Evaluation To evaluate the effectiveness of the question-answering functionality that combines knowledge graphs with large models, this study employed a combination of objective and subjective evaluations. Objective evaluation measured the effectiveness of the question-answering process based on answer accuracy and similarity metrics, while subjective evaluation involved professional teachers evaluating the results. This combined approach yielded more comprehensive feedback.

[0062] Objective evaluation is conducted through two methods: multiple-choice question verification and essay-based question verification. The multiple-choice question verification requires constructing a dataset containing 150 questions, covering three difficulty levels: easy, medium, and hard, with 50 questions at each level. Questions of varying difficulty levels were obtained from the exam preparation section of a middle school history teaching platform. Junior high school history exam questions from Zhejiang Province were screened and compiled using difficulty classification labels and average score rates. The experimental results are shown in Table 4.1 below, which lists the accuracy rates of the three large models on questions of varying difficulty levels, with or without the knowledge graph. This is calculated as the number of correctly answered questions divided by the total number of questions, with the number of correctly answered questions shown in parentheses.

[0063] Table 4.1 Verification of objective multiple-choice questions , The results show that without knowledge graph integration, simple questions have limited discrimination for each model, and the performance of the three models on this type of questions is similar. When faced with medium-difficulty questions, the GLM-4-Plus and Qwen-Max models showed similar levels of performance, while the performance of ERNIE 4.0 Turbo was slightly insufficient. When dealing with difficult questions, the GLM-4-Plus model performed better than the other two. When combined with the knowledge graph, although there is little room for improvement in the accuracy rate of simple questions, the role of the knowledge graph becomes more obvious as the difficulty of the questions increases, especially for difficult questions. In summary, the application of knowledge graphs in multiple-choice tests does help improve the model's answering effect, and the combination with GLM-4-Plus and Qwen-Max is even more outstanding.

[0064] The results show that without the integration of the knowledge graph, the discrimination of simple questions for each model is limited, and the performance of the three models on this type of questions is similar. In the face of medium-difficulty questions, the two models of GLM-4-Plus and Qwen-Max show similar levels, while the performance of ERNIE 4.0 Turbo is slightly insufficient. When dealing with high-difficulty questions, the GLM-4-Plus model performs better than the other two. When combined with the knowledge graph, although there is little room for improvement in the accuracy of simple questions, as the difficulty of the questions increases, especially in high-difficulty questions, the role of the knowledge graph is more obvious. To sum up, the application of the knowledge graph in the multiple-choice test does help to improve the answering effect of the model, and the combination of GLM-4-Plus and Qwen-Max performs better.

[0065] For the verification part of the short-answer questions, a dataset containing 100 historical short-answer questions is required. These questions are sourced from the examination papers of various districts in Zhejiang Province. The collected questions clearly include the reference answers, and the content examined by the questions has strong objectivity and weak subjective components. To compare the similarity between the generated answers and the reference answers, BLEU scores and ROUGE scores are used for evaluation.

[0066] BLEU scores are usually used in machine translation tasks. First, the n-gram similarities between two texts are calculated separately, where n is an integer between 1 and 4. For each value of n, a set of similarity scores is calculated. The set of similarity scores is normalized and substituted into the BLEU formula to calculate the final BLEU score.

[0067] , Compared with BLEU scores that focus on precision, ROUGE scores focus more on recall, that is, the ability of the generated text to capture the information of the reference text, and are widely used in question-answering systems and text generation tasks. The calculation method of ROUGE-N is to compare the n-gram of the generated text with the n-gram of the reference text, calculate the overlapping part between the two, and then divide it by the total number of n-grams in the reference text. The calculation formula is as follows: , ROUGE-L evaluates similarity based on the longest common subsequence (LCS). It not only considers the matching of words but also the order and position of words, so it can better capture the structure and coherence of the text.

[0068] , Combining the characteristics of the above indicators, this study selects ROUGE-1, ROUGE-2, and ROUGE-L indicators in the ROUGE scores. The specific data is shown in Table 4.2 below.

[0069] Table 4.2 Objective Q&A Verification , The research results show that after the introduction of the knowledge graph, the Q&A similarity indicators all show an upward trend, and compared with the BLEU score, the improvement of the overall ROUGE series of indicators is more significant. Combining the experimental data and theoretical content analysis, since the BLEU score mainly focuses on evaluating the exact word matching between the generated answer and the reference answer, without considering the content coverage, and the BLEU-4 score used here takes into account longer n-gram matching, so the score is relatively low. In summary, it verifies the effectiveness of the knowledge graph in enhancing the Q&A function of junior high school history, and the combination of the GLM-4-Plus model and the knowledge graph shows slightly better performance than other models in various indicators.

[0070] In the subjective evaluation session, this study invited three experienced front-line junior high school history teachers to participate, tested 50 open-ended history questions, compared six answers for each question, and selected the most preferred answer. As shown in the table below, Table 4.3 is the statistical analysis of 150 selections of the three teachers, and the values in parentheses represent the percentage of that selection.

[0071] Table 4.3 Subjective Q&A Verification , It can be seen from the data in the table that the answers combined with the knowledge graph are significantly more selected in the subjective evaluation, further verifying the effectiveness of the Q&A combined with the knowledge graph. At the same time, the combination of the Qwen-Max model and the knowledge graph shows better answering ability when dealing with open-ended history Q&A tasks.

[0072] To sum up, in the objective evaluation session, the performance of the GLM-4-Plus model and the Qwen-Max model combined with the knowledge graph is relatively close in the multiple-choice question verification, and in the short-answer question verification, the performance of the GLM-4-Plus combined with the knowledge graph is slightly better than the other two. In the subjective evaluation, the combination of the Qwen-Max model and the knowledge graph shows better performance in the open-ended history Q&A task. Given that the difference between the two models in the objective verification is not significant, and the Qwen-Max model is significantly better than the GLM-4-Plus model in the subjective evaluation, this study chooses to adopt the Qwen-Max model combined with the knowledge graph in the implementation of the Q&A function of the subsequent junior high school history learning assistance system.

[0073] (1) User Management Module It is divided into two login methods: student login and administrator login. Student users can fill in their real names, student IDs, classes and other information to register and log in. Teacher users can use the administrator account to manage students and can also edit the junior high school history knowledge graph and the content of the knowledge base.

[0074] After student users log in, click on the avatar in the upper right corner to enter the personal information page, where they can view and edit some personal information, and can also view their note history in the knowledge base for reviewing the key points of different history courses. After teacher users log in to the administrator account, in addition to being able to manage student accounts, they can also click on the details to view the note history of students to understand the learning progress of students in a timely manner. (2)Knowledge Graph Display Module The home page of the junior high school history learning assistance system presents the global structure of the knowledge graph. In the knowledge graph network, each node represents a historical knowledge point, and the connections between nodes represent the complex relationships between them.

[0075] (3)Knowledge Graph Query and Editing Module In the left menu bar, click on the expansion button to view the complete function navigation. Student users can use the sub-graph query function under the historical knowledge query module. After entering the historical knowledge point they want to retrieve, a sub-graph centered on this knowledge point and its detailed attributes will be displayed below. Enter the historical knowledge point "Zhang Qian", and it will show the relationships between entities of different categories including historical events, historical documents, historical groups, politics, and cultural creations and the historical figure "Zhang Qian", and display the attribute values such as alias / title, date of birth, date of death, achievements, positions, resource links, and learning objectives of "Zhang Qian" on the right.

[0076] The knowledge sub-graph can also be adjusted according to user needs. Configure restore, magicType, and saveAsImage in the toolbox to respectively implement the three functions of restoring the chart to the initial state, switching between the force-directed graph (force) and the chord graph (chord), and saving the chart as a picture.

[0077] Student users browse or query the detailed information according to the knowledge point category and name in the historical knowledge query module. Compared with sub-graph retrieval, if students are more concerned about mastering specific knowledge points rather than the associations between knowledge, the knowledge point query function will be more efficient and can appropriately reduce the cognitive burden of students. Teachers or administrators can, in the knowledge point query and editing section under the query and editing module classification, select the corresponding knowledge point type and then enter the knowledge point name for precise retrieval, or directly perform operations such as adding, editing, and deleting knowledge points under the knowledge point classification.

[0078] Student users input the starting knowledge point and the ending knowledge point in the relationship query section, and select the type of relationship to implement the query of knowledge point relationships. Teachers or administrator users can, on this basis, perform addition, editing, and deletion operations.

[0079] (4) Historical Q&A Module In the historical Q&A module, users can have conversations with the junior high school history Q&A assistant, and fill in their own knowledge gaps and thinking perspectives from the replied knowledge suggestions, so as to achieve the effect of learning assistance. The core of this function is based on the Q&A model that integrates the large model and the knowledge graph in Chapter 5. Through the powerful semantic understanding ability of the large model, combined with the professional historical knowledge and structure in the knowledge graph, it provides users with a more abundant and textbook-content-tailored answer generation method. Specifically, on the front end, the Vue.js framework is used to capture the questions input by users and serialize them into JSON objects, and then a request is sent to the back-end server. The request is distributed to the intelligent Q&A corresponding interface through routing, and processes such as keyword extraction, knowledge graph retrieval, and answer generation are executed. Finally, the function encapsulates the generated answers into a JSON response and returns them to the front end. The operations of the administrator in the editing module will also be synchronized to the knowledge graph used by the Q&A function, which not only improves the timeliness, accuracy, and relevance of the answers, but also enhances the learning experience of users. At the same time, on the right side of the interface, subgraphs closely related to the question will also be presented to promote users' understanding in the form of a combination of pictures and texts.

[0080] (5) Knowledge Base Module Student users can select the corresponding grade in the knowledge base learning module and learn and read resources such as course videos, documents, and courseware included therein. On the resource details page, the specific resource content is displayed on the left, and the knowledge points related to this course and the note input column are displayed on the right respectively. Student users can click on the knowledge points to enter the subgraph retrieval page to view the details. Through the association between resources and knowledge points, it helps students clarify the knowledge content and context of this lesson and make up for the deficiencies in historical learning. At the same time, student users can also record their learning gains in the form of notes and view the note history record in the personal interface.

[0081] Teachers or administrator users can upload course resources according to unit classification in the knowledge base management module, add and edit relevant knowledge points on the resource details page, and make timely adjustments according to the actual teaching situation.

[0082] The above is only the preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and all should be covered within the protection scope of the present invention.

Claims

1. A junior high school history auxiliary learning method based on a knowledge graph, characterized in that The method includes: Obtaining junior high school history teaching data; the sources of the junior high school history teaching data include: junior high school history textbooks, test question books, educational resource platforms, and knowledge bases; Constructing a pattern layer of a junior high school history knowledge graph based on the junior high school history teaching data; Preprocessing the junior high school history teaching data; Constructing a data layer of a junior high school history knowledge graph based on the preprocessed junior high school history teaching data and the pattern layer of the junior high school history knowledge graph; Assisting students in learning based on the data layer of the junior high school history knowledge graph.

2. The junior high school history assisted learning method based on a knowledge graph according to claim 1, wherein The specific process of constructing the pattern layer of the junior high school history knowledge graph based on the junior high school history teaching data is as follows: Sorting out the historical timeline based on the junior high school history teaching data, abstracting historical figures, historical events, geographical concepts, and political and economic system concepts, and classifying them into ancient, modern, and contemporary; Inviting front-line junior high school history teachers to adjust and summarize 11 categories of junior high school history concepts in combination with curriculum standard interpretation, teaching practice, and the five core historical qualities; Defining object attributes and data attributes for the junior high school history concepts, and constructing a pattern layer of the junior high school history knowledge graph.

3. The junior high school history assisted learning method based on a knowledge graph according to claim 2, characterized in that, The 11 categories of junior high school history concepts specifically include: Historical periods, historical figures, historical events, historical regions, historical groups, politics, economy, cultural creations, historical documents, art works, and ideological theories.

4. The junior high school history assisted learning method based on a knowledge graph according to claim 2, characterized in that The pattern layer of the junior high school history knowledge graph specifically includes: The object attribute system is used to express the relationships between different entities, including: there is a "successor" relationship between historical periods; there is a "being in" relationship between historical figures and historical periods; there are "kinship, related to" relationships between historical figures; there are "participating in, being involved in" relationships between historical figures and historical events; there is a "belonging to" relationship between historical figures and historical groups; there is a "being in" relationship between historical events and historical periods; there is a "taking place in" relationship between historical events and historical regions; there is an "including" relationship between historical events and politics, economy; there are "including, related to" relationships between historical events; there is a "participating in" relationship between historical groups and historical events; there is a "existing in" relationship between historical groups and historical periods; there is a "being located in" relationship between historical groups and historical regions; there is a "creator" relationship between civilization creations and historical figures; there is a "development" relationship between civilization creations; there is a "political association" relationship between politics; there is a "correlation" relationship between politics and economy; there are "related personnel" relationships between politics and historical figures; there is a "related organization" relationship between politics and historical groups; there is an "economic association" relationship between economy; there are "related personnel" relationships between economy and historical figures; there is a "related organization" relationship between economy and historical groups; there is an "introducing" relationship between historical documents and civilization creations; there is an "author" relationship between historical documents and historical figures; there is a "reflecting" relationship between historical documents and ideological theories; there is a "creator" relationship between art works and historical figures; there is a "creative prototype" relationship between art works and historical events; there is a "proposer" relationship between ideological theories and historical figures; there is a "proposing group" relationship between ideological theories and historical groups; The data attribute system is used to describe various concepts in detail. Among them, the data attributes of historical periods include period overview, specific years, political system, capital city, territory, status, learning objectives, and resource links; the data attributes of historical figures include alias / title, date of birth, date of death, achievements, positions, works, learning objectives, and resource links; the data attributes of historical events include background, time, process, results, impacts, significance, learning objectives, and resource links; the data attributes of historical regions include location, description, learning objectives, and resource links; the data attributes of historical groups include existence period, description, location, representative figures, status, learning objectives, and resource links; the data attributes of politics include period, creator / creating organization, concept, manifestation, role, learning objectives, and resource links; the data attributes of economy include period, proposer / proposing organization, concept, manifestation, role, learning objectives, and resource links; the data attributes of civilization creation include construction time / invention time, constructor / inventor, description, significance, learning objectives, and resource links; the data attributes of historical documents include author, time, brief introduction, content, significance, learning objectives, and resource links; the data attributes of art works include creator, time, brief introduction, significance, learning objectives, and resource links; the data attributes of ideological theories include proposed time, proposer, generalization, significance, evaluation, learning objectives, and resource links.

5. The junior high school history assisted learning method based on a knowledge graph according to claim 1, wherein, The preprocessing of the junior high school history teaching data is specifically as follows: For the junior high school history teaching data, highly homogenized and low-quality content is removed in combination with expert opinions; then it is converted into text information. The text information is normalized by removing punctuation marks that contribute nothing to semantic understanding, cleaning special characters, blank lines, and line breaks. After cleaning, the text is segmented, and the generated sentence list is converted into a format with one sentence per line to generate a TXT text.

6. The junior high school history assisted learning method based on a knowledge graph according to claim 1, characterized in that Based on the preprocessed junior high school history teaching data and the junior high school history knowledge graph schema layer, constructing the junior high school history knowledge graph data layer is specifically as follows: Entity extraction, relationship extraction, and attribute extraction are performed on the preprocessed text data to convert the junior high school history teaching data into structured knowledge. Among them, entity extraction is based on the schema layer annotation training model, relationship extraction adopts a combination of model and manual methods, and attribute extraction selects encyclopedia acquisition, large model extraction, or manual addition according to different attribute characteristics. A method combining attribute similarity and structural similarity is used for knowledge fusion. Potential duplicate entities are screened out by calculating similarity, and after manual verification, redundancy and conflicts are eliminated. The fused knowledge is stored in the Neo4j graph database, with historical knowledge entities represented by nodes and relationships between entities represented by edges to construct the junior high school history knowledge graph data layer.

7. The junior high school history assisted learning method based on a knowledge graph according to claim 1, characterized in that, Assisting students in learning based on the junior high school history knowledge graph data layer is specifically as follows: For the questions input by students, first, a domain determination is made to judge whether it belongs to the category of the history subject. If it does, the question keywords are extracted, relevant information is retrieved in the knowledge graph, and the obtained information is optimized after retrieval to remove redundant content. Integrate the optimized information with the original question. If there is a conversation history, extract the key information from the conversation history and combine it with the original question to form a comprehensive prompt template that is input into the large model to generate the answer.

8. A junior high school history assisted learning system based on a knowledge graph, characterized in that, The system comprises: Data acquisition module: responsible for obtaining junior high school history teaching data from junior high school history textbooks, test books, educational resource platforms and knowledge bases; Model layer construction module: Based on the acquired junior high school history teaching data, the historical timeline is sorted out, historical concepts are abstracted and classified, teachers are invited to adjust and summarize the concept system, object attributes and data attributes are defined, and the model layer of the junior high school history knowledge graph is constructed; Data preprocessing module: processes the acquired junior high school history teaching data, removes highly homogeneous and low-quality content, converts data in various formats into text information, and then normalizes the text information to generate TXT text for subsequent use; Data layer construction module: Based on the pre-processed data and model layer, it performs knowledge extraction, knowledge fusion and knowledge storage to build the junior high school history knowledge graph data layer; Learning assistance module: Based on the junior high school history knowledge graph data layer, it realizes the knowledge question and answer function.

9. The junior high school history assisted learning system based on a knowledge graph according to claim 8, wherein, The learning assistance module includes: User management module: supports student and teacher login; students use it to complete information and view notes; teachers use it to manage student accounts, view learning progress, and edit knowledge graphs and knowledge bases; Knowledge graph display module: Visually presents the knowledge graph, allowing users to zoom in and out, drag and browse the entire graph, filter specific types of knowledge points through legends, and view the knowledge context; Knowledge graph query module: students conduct subgraph and knowledge point queries; Knowledge graph editing module: teachers or administrators can add, delete, and modify knowledge points and relationships in the knowledge graph; Knowledge Base Module: Stores learning resources by grade. Students select their grade to study, view resources and related knowledge points, and take notes. Teachers or administrators upload resources and edit knowledge points. Historical question-answering module: combines knowledge graphs and large models to determine the problem domain, retrieve information and generate answers.

Images