A knowledge graph-based teaching quality intelligent evaluation method and system

Abstract

The application relates to the technical field of teaching evaluation, and discloses a teaching quality intelligent evaluation method and system based on a knowledge graph, which comprises the following steps: constructing a subject basic knowledge graph, and generating a time sequence teaching event set; according to a preset classroom time driving update rule, identifying a teaching trigger event in the time sequence teaching event set, and generating a time sequence teaching knowledge subgraph evolving with a teaching process; according to entity features and correlation relationship features in the time sequence teaching knowledge subgraph, generating a graph feature set, mapping the graph feature set to a teaching quality evaluation dimension, combining a teaching business constraint, and calculating a teaching quality intermediate evaluation result; mapping the teaching quality intermediate evaluation result back to the time sequence teaching knowledge subgraph, obtaining a five-level traceable evidence chain, generating a visual evaluation evidence graph, and displaying a deduction path and a basis of each evaluation conclusion, so that a dynamic intelligent evaluation of the teaching evaluation process is realized, and the whole process from original events to evaluation conclusions is traceable and interpretable.

Description

Technical Field

[0001] This invention relates to the field of teaching evaluation technology, and in particular to a knowledge graph-based intelligent evaluation method and system for teaching quality. Background Technology

[0002] With the deep integration of artificial intelligence and educational informatization, intelligent teaching evaluation technology based on knowledge graphs has become an important research direction in the field of educational technology. Currently, some researchers have proposed methods to construct subject knowledge systems using knowledge graphs. By organizing course knowledge points into a structured network, this provides basic support for the organization of teaching resources and the planning of learning paths. In terms of classroom teaching evaluation, existing systems can acquire multimodal classroom data through audio and video acquisition devices and record and analyze the teaching process by combining technologies such as speech recognition and image recognition. For example, they can conduct dual-dimensional analysis of teacher behavior and student behavior to generate classroom participation reports and knowledge point slices. In addition, some studies have attempted to apply knowledge graphs to teaching quality evaluation by constructing teaching knowledge graphs and comparing the actual teaching path with the preset optimal path to achieve the evaluation function.

[0003] However, the aforementioned technologies still have shortcomings in terms of dynamism, interpretability, and traceability of evaluation conclusions. On the one hand, existing knowledge graph-based teaching evaluation methods mostly adopt static knowledge graph structures, which are difficult to reflect the dynamic evolution of knowledge status with teaching activities during the classroom teaching process. Classroom teaching is a temporal and dynamic process. The order of knowledge point explanation, the content of teacher-student interaction, and the shift of teaching focus all change with the classroom process. Static graphs cannot capture this temporal evolution characteristics, resulting in the evaluation results being out of touch with the actual classroom teaching process. On the other hand, most current intelligent evaluation systems use black-box models to output evaluation scores, and the evaluation conclusions lack traceable derivation basis. Teachers and administrators find it difficult to understand the logic behind the evaluation results and cannot locate the problems in specific teaching links. These shortcomings limit the practical application value of intelligent evaluation technology in improving classroom teaching. Summary of the Invention

[0004] This invention provides a knowledge graph-based intelligent evaluation method and system for teaching quality to solve the problems mentioned in the background.

[0005] To achieve the above objectives, this invention provides an intelligent teaching quality assessment method based on knowledge graphs, comprising: S1: Construct a subject-based basic knowledge graph containing knowledge point entities and inherent semantic relationships between knowledge point entities, collect the temporal teaching event flow in the classroom teaching process, and generate a temporal teaching event set; S2: Based on the preset classroom time-driven update rules, identify teaching trigger events in the time-series teaching event set, modify the entity confidence and association weights within the subject basic knowledge graph according to the teaching trigger events, and generate a time-series teaching knowledge subgraph that evolves with the teaching process; S3: Based on the entity features and relational features within the time-series teaching knowledge subgraph, generate a graph feature set, map the graph feature set to the teaching quality assessment dimension, and calculate the intermediate teaching quality assessment results in conjunction with teaching business constraints. S4: Map the intermediate evaluation results of teaching quality back to the time-series teaching knowledge subgraph to obtain a five-level traceable evidence chain, and generate a visual evaluation evidence graph to show the derivation path and basis of each evaluation conclusion.

[0006] To address the aforementioned problems, this invention also provides a knowledge graph-based intelligent evaluation system for teaching quality, the system comprising: The graph construction and acquisition module is used to construct a subject-based basic knowledge graph containing knowledge point entities and the inherent semantic relationships between knowledge point entities, and to collect the temporal teaching event stream in the classroom teaching process to generate a temporal teaching event set. The event-driven update module is used to identify teaching trigger events in the set of time-series teaching events according to the preset class time-driven update rules, modify the confidence and association weights of entities within the subject basic knowledge graph according to the teaching trigger events, and generate time-series teaching knowledge subgraphs that evolve with the teaching process. The feature mapping evaluation module is used to generate a graph feature set based on the entity features and relationship features within the time-series teaching knowledge subgraph, map the graph feature set to the teaching quality evaluation dimension, and calculate the intermediate evaluation results of teaching quality in combination with teaching business constraints. The evidence visualization module is used to map intermediate evaluation results of teaching quality back to the time-series teaching knowledge subgraph, obtain a five-level traceable evidence chain, and generate a visualized evaluation evidence map to show the derivation path and basis of each evaluation conclusion.

[0007] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention establishes classroom time-driven update rules to identify teaching trigger events in a time-series teaching event set in real time. Based on the specific type and occurrence time of the teaching trigger events, it dynamically modifies the confidence values ​​of corresponding knowledge point entities and the weight values ​​of relationships between entities within the subject-based basic knowledge graph, generating a time-series teaching knowledge subgraph that evolves with the teaching process. This technique enables the knowledge graph to be updated in real time as classroom teaching activities progress. The importance confidence of knowledge points and the strength of relationships between knowledge points are dynamically adjusted according to the actual teaching progress and teacher-student interaction content. This solves the problem that traditional static knowledge graphs cannot reflect the time-series evolution characteristics of classroom teaching, ensuring that the knowledge graph used for subsequent evaluation remains synchronized with the current classroom progress.

[0008] 2. This invention maps intermediate teaching quality assessment results back to a temporal teaching knowledge subgraph, constructing a five-level progressive evidence sequence from the original teaching event records in the temporal teaching event set, teaching trigger event records, entity confidence and association weight records of the modified temporal teaching knowledge subgraph, graph feature set records, to the intermediate teaching quality assessment result records. Based on this evidence chain, the nodes and edges of the temporal teaching knowledge subgraph are rendered to generate a visual assessment evidence graph. This technique allows each assessment conclusion to be traced back to the original teaching event step by step along the five-level evidence chain. Teachers and administrators can intuitively view the complete derivation path of the assessment conclusion and the specific basis of each level through the visual graph, improving the interpretability of the assessment results and the operability of teaching improvement. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating a knowledge graph-based intelligent assessment method for teaching quality, provided as an embodiment of the present invention. Figure 2 This is a functional module diagram of a knowledge graph-based intelligent evaluation system for teaching quality, provided as an embodiment of the present invention.

[0010] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0011] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0012] This application provides a knowledge graph-based intelligent assessment method for teaching quality. The execution subject of this knowledge graph-based intelligent assessment method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the knowledge graph-based intelligent assessment method for teaching quality can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0013] Reference Figure 1 The diagram shown is a flowchart illustrating a knowledge graph-based intelligent assessment method for teaching quality according to an embodiment of the present invention. In this embodiment, the knowledge graph-based intelligent assessment method for teaching quality includes: S1: Construct a subject-based basic knowledge graph containing knowledge point entities and inherent semantic relationships between knowledge point entities, collect the temporal teaching event flow in the classroom teaching process, and generate a temporal teaching event set; In this embodiment of the invention, the core concept system of the target discipline is determined based on the standardized subject corpus of the target discipline, a class hierarchy structure of knowledge point entities is established, and a subject knowledge ontology is generated. It should be specifically explained that the first step is to acquire the teaching resource data of the target subject. This teaching resource data includes the textbook text, course syllabus, teaching courseware, and exercise set of the subject. This teaching resource data is then cleaned and standardized. The word segmentation dictionary consists of professional terms and commonly used words in the subject area and can be constructed by those skilled in the art based on the specific subject content.

[0014] Based on a standardized subject corpus, the core concept system of the target subject is determined. The method for determining the core concept system is as follows: the frequency of occurrence of each term in the standardized subject corpus is statistically analyzed, and terms with a frequency higher than a preset frequency threshold are taken as core concepts. The value of the preset frequency threshold can be set by a person skilled in the art based on the size of the corpus and the granularity of analysis. For example, when the total number of words in the corpus is about 100,000, the threshold can be set to 10. After determining the core concept system, a hierarchical structure of knowledge point entities is established based on the genus-species relationship, whole-part relationship, and sequential relationship between the core concepts. The genus-species relationship refers to the fact that one knowledge point is a subclass of another knowledge point.

[0015] Based on the subject knowledge ontology, knowledge point entities are extracted from the standardized subject corpus to obtain a set of knowledge point entities. Based on the subject knowledge ontology and the set of knowledge point entities, semantic relationships between knowledge point entities are extracted from the standardized subject corpus to obtain a set of entity relationships. It should be noted that the named entity recognition method uses a sequence labeling method based on conditional random fields. This method is trained in advance using labeled subject corpora and can identify specific words belonging to knowledge point entity types from the text.

[0016] For each pair of knowledge point entities in the knowledge point entity set, a co-occurrence text window containing the pair of knowledge point entities is retrieved in the standardized subject corpus. The size of the co-occurrence text window is a text fragment formed by expanding a preset number of characters or words to the left and right of the position where each knowledge point entity appears. In this embodiment, the preset number of expanded characters is two hundred characters.

[0017] Within the co-occurrence text window, a relation extraction method based on dependency parsing is used to determine the semantic association type between the pair of knowledge point entities. Dependency parsing analyzes the grammatical structure of sentences to determine the subject-verb-object, attributive, adverbial, and complement dependencies between words, thereby identifying whether there is an association between knowledge point entities and the specific type of such association. Association types include, but are not limited to: predecessor-successor relationship, i.e., one knowledge point precedes another in the cognitive order; inclusion relationship, i.e., one knowledge point contains another knowledge point as its sub-content; equivalence relationship, i.e., two knowledge points express the same or similar content; and relevance relationship, i.e., there is no explicit order or inclusion relationship between two knowledge points but there is a contextual relationship. After extracting the relations of all entity pairs in the knowledge point entity set, an entity relation set is obtained. Each entity relation record in this set contains the identifiers of the two knowledge point entities and the association type between them.

[0018] Based on the set of knowledge point entities and the set of entity relationships, a subject-based knowledge graph is constructed, with knowledge point entities as nodes and semantic relationships as edges.

[0019] Specifically, each knowledge point entity in the knowledge point entity set is treated as a node in the graph. The node's identification information includes the name of the knowledge point entity, the type of knowledge point entity it belongs to, and its hierarchical position in the subject knowledge ontology. Each semantic relationship in the entity relationship set is treated as an edge in the graph. The edge's identification information includes the identifiers of the two knowledge point entities it connects and the relationship type. When constructing the graph, a graph database is used as the storage medium. Nodes and edges are written into the graph database according to the graph data structure. The method of writing nodes and edges into the graph database is as follows: first, all nodes are created, and each node has a unique node identifier; then, all edges are created, and each edge records its starting node identifier, ending node identifier, and relationship type. Through the above method, the construction of the subject basic knowledge graph is completed.

[0020] Multi-source event detection is performed on the original audio and video streams during classroom teaching to obtain multimodal teaching original event records, including teacher speech segments, student speech segments, blackboard content frames, and screen projection content frames; It should be specifically noted that audio and video acquisition equipment is deployed in the classroom. This equipment includes a high-definition camera installed at the front of the classroom and a directional microphone array installed in the podium area. The high-definition camera continuously acquires the original audio and video stream of the classroom at a preset sampling frequency, which can be 30 frames per second. The directional microphone array is used to collect the teacher's voice in a directional manner. Its directionality allows it to mainly pick up the sound source in the podium area, reducing the interference of classroom environmental noise.

[0021] For the acquired raw audio and video streams, the audio and video streams are first processed separately to obtain the raw audio and raw video streams. Speech activity detection is then performed on the raw audio stream. Speech activity detection determines whether the current audio frame contains a speech signal by analyzing the short-time energy and zero-crossing rate of the audio frame. When the short-time energy is higher than a preset energy threshold and the zero-crossing rate is within a preset range, the current audio frame is determined to contain a speech signal. After detecting the speech signal, speaker identification is further performed. Speaker identification is based on a pre-trained speaker voiceprint model. This model has pre-collected the voiceprint features of the teacher and students in the class and established a voiceprint database. The audio frame containing the speech signal is input into the speaker voiceprint model, and its voiceprint features are extracted and compared with the voiceprint templates in the voiceprint database to determine whether the speech signal belongs to the teacher or the student. Based on the speaker identification results, the raw audio stream is divided into teacher speech segments and student speech segments to form the audio part of the multimodal teaching raw event record.

[0022] For the original video stream, whiteboard area detection and screen projection area detection are performed. Whiteboard area detection is achieved by identifying the boundaries of the blackboard or whiteboard area in the video frame and extracting the pixel content within that area. Screen projection area detection is achieved by identifying the display area of ​​the projection screen or electronic screen in the video frame and extracting the pixel content within that area. After the whiteboard area detection and screen projection area detection are completed, whiteboard content frames and screen projection content frames are extracted from the original video stream according to the timestamps to form the video part of the multimodal teaching original event record. The above audio part and video part are collectively referred to as the multimodal teaching original event record.

[0023] The teacher's speech segment and the student's speech segment are converted into text to obtain the teacher's speech text and the student's speech text. The blackboard content frame is subjected to optical character recognition to obtain the blackboard text. The screen projection content frame is parsed to obtain the projection text. It should be specifically noted that the speech-to-text conversion uses a deep learning-based automatic speech recognition model. This model employs an end-to-end encoder-decoder architecture. The encoder converts the input audio signal into an acoustic feature sequence, and the decoder generates the corresponding text sequence based on this acoustic feature sequence. For teacher and student speech segments, they are input into the automatic speech recognition model, and the model outputs the corresponding text sequences, which serve as the teacher's and student's speech texts, respectively. During the above processing, the output results of the speech-to-text conversion are scored for confidence. When the confidence of a word or segment is lower than a preset confidence threshold, a confidence label is added to the output text for that word or segment. This processing enables subsequent steps to recognize and utilize confidence information for text alignment and event detection.

[0024] Optical character recognition (OCR) employs a scene text recognition model based on a combination of convolutional neural networks (CNNs) and recurrent neural networks (RNNs). This model first extracts visual feature maps from the blackboard content frame using a CNN, then uses a RNN to serialize and decode the visual feature maps, recognizing the character sequences in the image. For handwritten text in the blackboard content frame, the OCR output is text content that has undergone stroke correction and character standardization. For formulas, special symbols, and charts contained in the blackboard content frame, the OCR model outputs corresponding text descriptions or structured representations. After OCR is completed, all the text recognized in the same blackboard content frame is arranged according to its spatial position in the image to generate the blackboard text.

[0025] Content parsing first determines the type of the screen-projected content frame, which can be a presentation page, a document page, or a webpage. This type determination is achieved by analyzing the color distribution and layout characteristics of the screen-projected content frame. For example, presentation pages typically have a hierarchical structure with title and body text areas. Based on the determined type, the corresponding content parser is invoked for parsing. For presentation pages, the title and body text are extracted, preserving the hierarchical structure. For document pages, continuous text content is extracted. For webpages, the main text content is extracted, filtering out non-main content such as navigation bars and advertising areas. The output of content parsing is structured projected text, which includes the text content and its position information within the source page.

[0026] The teacher's spoken text, student's spoken text, blackboard text, and projected text are aligned according to timestamps to generate a teaching text sequence. It needs to be specifically explained that the timestamp information of the teacher's speech segments, student's speech segments, blackboard content frames, and screen projection content frames are obtained from the original event records of multimodal teaching. The timestamps of the teacher's speech text and student's speech text correspond to the time range of their respective original audio segments; the timestamps of the blackboard text correspond to the acquisition time of the blackboard content frames; and the timestamps of the projection text correspond to the acquisition time of the screen projection content frames. The teacher's speech text, student's speech text, blackboard text, and projection text are mapped onto a unified timeline according to their respective timestamps, so that text content at the same point in time or within the same time range can be associated. After the timestamps are aligned, the four types of text are spliced ​​together in chronological order, and multiple types of text within the same time range are merged and recorded to generate a teaching text sequence.

[0027] Semantic event boundary monitoring is performed on the teaching text sequence to obtain semantic turning points. The teaching text sequence is then segmented based on the semantic turning points to obtain teaching event units. It should be specifically explained that semantic event boundary detection is achieved by analyzing the content coherence and topic switching features in the teaching text sequence. First, the teaching text sequence is divided into sentence sequences at the sentence level. For each pair of adjacent sentences in the sentence sequence, the semantic similarity between the two sentences is calculated. The semantic similarity calculation adopts a sentence vector encoding method based on a pre-trained language model, encoding each sentence into a fixed-length semantic vector. Then, the cosine similarity between the two semantic vectors is calculated. When the cosine similarity is lower than a preset similarity threshold, the boundary position of the two adjacent sentences is marked as a semantic inflection point. The preset similarity threshold can be set by those skilled in the art within the range of 0.3 to 0.5 according to the coherence features of the specific teaching content. After determining all semantic inflection points, the teaching text sequence is divided into several continuous text segments using the semantic inflection points as the segmentation positions. Each text segment constitutes a teaching event unit, and each teaching event unit represents a teaching segment with relatively complete semantics and a relatively focused topic.

[0028] Add event timestamps, event type tags, and event participant identifiers to the teaching event units, and combine the teaching event units in chronological order to obtain a set of time-series teaching events.

[0029] It should be specifically noted that for each teaching event unit, the start and end times of the teaching event unit are determined based on the timestamp information of all the text content constituting that teaching event unit. The start and end times are used as the event timestamps of that teaching event unit. Based on the content characteristics of the teaching event unit, its event type label is automatically determined. Event type labels include, but are not limited to: lecture, which refers to the teacher explaining knowledge points; question and answer, which refers to the interaction between the teacher and students asking and answering questions; demonstration, which refers to the teacher giving a teaching demonstration or operation demonstration; and practice, which refers to students doing classroom exercises or solving problems.

[0030] Event type labeling is determined using a combination of keyword matching and text classification models. Keyword matching identifies obvious type indicators; for example, if a teaching event unit contains keywords like "question" or "answer," it is classified as a question-and-answer event. The text classification model handles cases without obvious keywords. Event participant identification is determined by statistically analyzing the occurrence of teacher and student statements within the teaching event unit. When a teaching event unit contains only teacher statements, the participant is identified as a teacher; when it contains only student statements, the participant is identified as a student; and when it contains both teacher and student statements, the participant is identified as both a teacher and a student. After adding this information to each teaching event unit, all teaching event units are arranged according to their timestamps to generate a chronological teaching event set. Each teaching event unit in this chronological teaching event set is arranged chronologically, forming an event flow structure for the classroom teaching process, serving as input for subsequent steps.

[0031] S2: Based on the preset classroom time-driven update rules, identify teaching trigger events in the time-series teaching event set, modify the entity confidence and association weights within the subject basic knowledge graph according to the teaching trigger events, and generate a time-series teaching knowledge subgraph that evolves with the teaching process; In this embodiment of the invention, the time-series teaching event set is divided into a continuous time window sequence according to the order of the event timestamps of each teaching event unit in the time-series teaching event set. It is important to explain that the process begins by retrieving the event timestamp of each teaching event unit from the chronological teaching event set. Each event timestamp contains the start and end times of the teaching event unit, recorded in seconds with the start of the class as the time origin. All teaching event units are arranged into a chronologically ordered sequence according to the order of their timestamps. Then, based on a preset time window width, this event sequence is divided into several consecutive time windows. The time window width represents the length of time covered by each time window, and this width can be set by those skilled in the art based on the actual duration of the class and the granularity of the evaluation. For example, when the total class duration is forty-five minutes, the time window width can be set to three minutes, thus dividing the entire class into fifteen consecutive time windows. Each time window contains several teaching event units, and the event timestamps of these teaching event units all fall within the time range of that time window. The above division results are then arranged in chronological order to obtain a continuous sequence of time windows.

[0032] Based on the preset classroom time-driven update rules, the event type labels and event participant identifiers of the teaching event units within each time window are extracted from the time window sequence, and the event type distribution vector and participant distribution vector of the current window are generated. Specifically, for each time window in the time window sequence, all teaching event units within that time window are traversed, and the event type label and event participant identifier for each teaching event unit are read. The set of values ​​for the event type label is a predefined finite set, which includes four event types: lecture, Q&A, demonstration, and practice. The set of values ​​for the event participant identifier includes two types of identifiers: teacher and student. For a given time window, the frequency of each event type label within that window is counted, resulting in a four-dimensional count vector. Each dimension of this vector corresponds to one event type label, and the value of the dimension is the frequency of that event type within the time window. This four-dimensional count vector is used as the event type distribution vector for the current window. The frequency of each event participant identifier within that window is counted, resulting in a two-dimensional count vector. Each dimension of this vector corresponds to one event participant identifier, and the value of the dimension is the frequency of that participant identifier within the time window. This two-dimensional count vector is used as the participant distribution vector for the current window. For each time window, the corresponding event type distribution vector and participant distribution vector for the current window are generated in the above manner.

[0033] The event type distribution vector and participant distribution vector of the current window are compared with the historical event type distribution vector and historical participant distribution vector of the previous time window to obtain the change in event type and the change in participant distribution. Specifically, for the event type distribution vector of the current time window, a vector subtraction operation is performed between the event type distribution vector of the current window and the event type distribution vector of the previous time window. That is, the count value of each dimension of the current window is subtracted from the count value of the corresponding dimension of the previous time window to obtain the event type change. This event type change is a four-dimensional vector, and the value of each dimension is positive, negative, or zero. A positive value indicates that the number of times the event type appears in the current window has increased compared to the previous window, a negative value indicates that it has decreased, and zero indicates that there is no change. For the participant distribution vector of the current time window, a vector subtraction operation is performed between the participant distribution vector of the current window and the participant distribution vector of the previous time window to obtain the participant distribution change. This participant distribution change is a two-dimensional vector, and the value of each dimension is positive, negative, or zero, representing the increase or decrease in the number of times teachers or students participate in the window, respectively.

[0034] Identify teaching-triggered events based on changes in event type and participant distribution, including: The changes in event types and participant distribution are analyzed to obtain the sets of event type addition identifiers, event type disappearance identifiers, participant addition identifiers, and participant disappearance identifiers. It is necessary to explain in detail the process of parsing the event type change, obtaining the specific dimension values ​​of the event type change, and checking each dimension of the event type change. When a dimension has a positive value, it means that the event type appears for the first time in the current window, and the event type represented by the dimension corresponding to the positive value is recorded as a new event type. When a dimension has a negative value, it means that the event type has decreased in the current window compared to the previous window, and the event type represented by the dimension corresponding to the negative value is recorded as a decreased event type. All new event types are collected into a set of event type new identifiers, and all decreased event types are collected into a set of event type disappearance identifiers.

[0035] Specifically, when an event type appears in the current window but not in the previous window, it is classified as a newly added event type; when an event type appears in the previous window but not in the current window, it is classified as a disappeared event type. Similarly, the change in participant distribution is analyzed to obtain specific dimension values. When the teacher dimension value is positive, the teacher identifier is added to the participant new identifier set; when the teacher dimension value is negative, the teacher identifier is added to the participant disappearance identifier set. When the student dimension value is positive, the student identifier is added to the participant new identifier set; when the student dimension value is negative, the student identifier is added to the participant disappearance identifier set.

[0036] Based on the set of newly added event type identifiers and the set of event type disappearance identifiers, the first occurrence time and the last occurrence time of the event type within the current time window are marked as the event triggered by the addition of the event type and the event triggered by the disappearance of the event type, respectively. Specifically, for each newly added event type in the event type addition identifier set, all corresponding teaching event units are searched within the current time window. The teaching event unit with the earliest event timestamp is selected, and its event timestamp is taken as the first appearance time of the event type within the current window. This teaching event unit is then marked as a type addition trigger event. For each disappeared event type in the event type disappearance identifier set, all corresponding teaching event units are searched within the current time window. The teaching event unit with the latest event timestamp is selected, and its event timestamp is taken as the last appearance time of the event type within the current window. This teaching event unit is then marked as a type disappearance trigger event. A type addition trigger event indicates the start of a new teaching activity at that moment, while a type disappearance trigger event indicates the end of a teaching activity at that moment.

[0037] Based on the set of participant addition identifiers and the set of participant disappearance identifiers, the first time a participant participates and the last time a participant participates within the current time window are marked as the participant addition trigger event and the participant disappearance trigger event, respectively. Specifically, for each newly added participant identifier in the set of newly added participant identifiers, all corresponding teaching event units are searched within the current time window. The teaching event unit with the earliest event timestamp is selected, and its event timestamp is taken as the first participation time of the participant identifier within the current window. This teaching event unit is then marked as the participant addition trigger event. Similarly, for each disappeared participant identifier in the set of disappeared participant identifiers, all corresponding teaching event units are searched within the current time window. The teaching event unit with the latest event timestamp is selected, and its event timestamp is taken as the last participation time of the participant identifier within the current window. This teaching event unit is then marked as the participant disappearance trigger event. A participant addition trigger event indicates that the participant role entered for the first time at that moment, while a participant disappearance trigger event indicates that the participant role exited at that moment.

[0038] Summarize all new trigger events of all types, trigger events for the disappearance of types, trigger events for the addition of participants, and trigger events for the disappearance of participants to generate a list of teaching trigger events.

[0039] Specifically, all marked new, disappearance, participant addition, and participant disappearance trigger events within the current time window are gathered together and sorted according to their respective event timestamps. After sorting, a trigger event type label is added to each trigger event. This label is used to distinguish whether the trigger event is a new type, disappearance type, participant addition, or participant disappearance. The sorted and labeled trigger event records are summarized to form a teaching trigger event list for the current time window. Each entry in this teaching trigger event list includes the occurrence time of the trigger event, the trigger type or participant, and the trigger category. For each time window, a corresponding teaching trigger event list is generated in the above manner, and the teaching trigger event lists of each time window are passed to subsequent steps in chronological order.

[0040] Based on the sequence of trigger event types in the list of teaching trigger events, determine the set of target knowledge point entities and the set of target association relationships corresponding to each teaching trigger event in the subject basic knowledge graph; Specifically, the process involves obtaining the list of teaching trigger events for the current time window and extracting the trigger event type label for each event. Based on the event type information in the label, the type of teaching activity involved is determined. For example, if the trigger event type label includes "lecture," it indicates that the associated knowledge point is related to lecture content. Based on this teaching activity type, the corresponding list of knowledge point entities is retrieved from a pre-defined mapping table of the subject knowledge ontology. This mapping table predefines the correspondence between each teaching activity type and knowledge point entities in the subject's basic knowledge graph. All retrieved knowledge point entities are aggregated to form a target knowledge point entity set. For each knowledge point entity in the target knowledge point entity set, all connections between that entity and the subject's basic knowledge graph are obtained. These connections are then aggregated to form a target association set. In this way, each teaching trigger event is mapped to a specific knowledge point entity and its associated relationships in the subject's basic knowledge graph.

[0041] Based on the trigger time sequence in the teaching trigger event list, perform a confidence increment operation on each target knowledge point entity in the target knowledge point entity set to generate an updated entity confidence value, and perform a weight increment operation on each relationship in the target relationship set to generate an updated relationship weight value. Specifically, for each target knowledge point entity in the target knowledge point entity set, the current entity confidence value is obtained. The entity confidence value represents the degree to which the knowledge point entity is mentioned or emphasized in the current classroom teaching process, and its value ranges from 0 to 1. Based on the occurrence time of the triggering event in the trigger time sequence, the increment magnitude of this confidence increment operation is determined. The increment magnitude is related to the time interval between the occurrence time of the triggering event and the start time of the current time window: the earlier the triggering event occurs, the smaller its impact on the current knowledge state, and the smaller the increment magnitude; the later the triggering event occurs, the larger its impact on the current knowledge state, and the larger the increment magnitude. The current entity confidence value is added to the calculated increment magnitude to obtain the updated entity confidence value. After performing the confidence increment operation, the updated entity confidence value is normalized, limiting its value to the range of 0 to 1 to ensure the consistency of the physical meaning of the confidence value.

[0042] For each relationship in the target set of relationships, obtain its current relationship weight value. The relationship weight value represents the strength of the connection between the two knowledge point entities connected by the relationship in the current classroom teaching process, and its value is a real number between 0 and 1. Based on the occurrence time of the triggering event in the trigger time sequence, determine the increment magnitude of this weight increment operation. The method for determining the increment magnitude is the same as that for the confidence increment operation, i.e., it is related to the time interval between the occurrence time of the triggering event and the start time of the current time window. Add the calculated increment magnitude to the current relationship weight value to obtain the updated relationship weight value. After performing the weight increment operation, normalize the updated relationship weight value, limiting its value to the range of 0 to 1 to ensure the consistency of the physical meaning of the weight value.

[0043] Update the subject-specific knowledge graph based on the new entity confidence value and the updated association weight value; Specifically, this involves reading all knowledge point entities and relationships stored in the subject-specific knowledge graph. For each target knowledge point entity, its original entity confidence value is replaced with the corresponding updated entity confidence value. For each target relationship, its original relationship weight value is replaced with the corresponding updated relationship weight value. For knowledge point entities and relationships not involved in the triggered event, their confidence and weight values ​​remain unchanged. After all replacement operations are completed, the attribute values ​​of nodes and edges related to the triggered event in the subject-specific knowledge graph are updated to the latest values ​​corresponding to the current time window, forming an updated subject-specific knowledge graph. This update operation enables the subject-specific knowledge graph to reflect the dynamic changes in the importance of knowledge points and the strength of relationships between knowledge points in the current classroom teaching process.

[0044] Using the current time point of the classroom teaching process as the cutoff boundary, a time-series teaching knowledge subgraph is generated based on the knowledge point entities and relationships related to the current classroom time period in the updated subject basic knowledge graph.

[0045] Specifically, it's necessary to clarify the time point for obtaining the current classroom teaching progress. This time point can be represented as the length of time from the start of the class to the current moment. Using this time point as the cutoff boundary, the currently completed class time segment is determined. Based on this class time segment, knowledge point entities are filtered from the updated subject-specific knowledge graph. The filtering criterion is whether the event timestamps of the teaching event units associated with each knowledge point entity fall within the currently completed class time segment. If a knowledge point entity has at least one associated teaching event unit whose event timestamp falls within this time segment, then that knowledge point entity is included in the temporal teaching knowledge subgraph. Similarly, for associations, if both knowledge point entities connected by a certain association are included in the temporal teaching knowledge subgraph, and this association exists in the updated subject-specific knowledge graph, then this association is included in the temporal teaching knowledge subgraph. After the filtering is completed, a subgraph is constructed using the included knowledge point entities as nodes and the included associations as edges; this subgraph is the temporal teaching knowledge subgraph. The node confidence values ​​and edge weight values ​​of this time-series teaching knowledge subgraph are all updated values, reflecting the knowledge status of the classroom teaching process up to the current moment.

[0046] S3: Based on the entity features and relational features within the time-series teaching knowledge subgraph, generate a graph feature set, map the graph feature set to the teaching quality assessment dimension, and calculate the intermediate teaching quality assessment results in conjunction with teaching business constraints. In this embodiment of the invention, all nodes of the time-series teaching knowledge subgraph are traversed, and a set of node feature records is generated based on the entity confidence value of the knowledge point entity corresponding to each node and the knowledge point level category to which the entity belongs. It is important to first obtain all nodes contained in the temporal teaching knowledge subgraph. Each node in the temporal teaching knowledge subgraph corresponds to a knowledge point entity, which has a defined knowledge point hierarchy category within the subject knowledge ontology. The knowledge point hierarchy category indicates the hierarchical position of the knowledge point entity in the subject knowledge system, including but not limited to: the basic concept layer, indicating that the knowledge point is the basic concept and definition of the subject; the core principle layer, indicating that the knowledge point is the core theory and basic principle of the subject; and the application method layer, indicating that the knowledge point is the specific application method and operational skill of the subject knowledge. When traversing each node, the entity confidence value of the knowledge point entity corresponding to that node is read. The entity confidence value is a real number between 0 and 1, representing the degree of mention or importance of the knowledge point entity in the current classroom teaching process. At the same time, the knowledge point hierarchy category of the knowledge point entity corresponding to that node is read. For each node, the node identifier, entity confidence value, and knowledge point hierarchy category are combined into a node feature record. All node feature records obtained after traversing all nodes are collected to generate a node feature record set.

[0047] Traverse all edges of the temporal teaching knowledge subgraph, and generate an edge feature set based on the association weight value and association type of the semantic association relationship corresponding to each edge; It's important to clarify that each edge in the temporal teaching knowledge subgraph corresponds to a semantic association between two knowledge point entities. Semantic associations have distinct association types, including predecessor-successor, inclusion, equivalence, and relevance. A predecessor-successor relationship indicates a sequential dependency between two knowledge points, such as one knowledge point needing to be mastered before another. An inclusion relationship indicates that one knowledge point contains another as its sub-content. An equivalence relationship indicates that two knowledge points express the same or similar meaning in terms of content. A relevance relationship indicates that there is no explicit order or inclusion relationship between two knowledge points, but there is a contextual connection. When traversing each edge, the association weight value of the corresponding semantic association is read. This association weight value is a real number between 0 and 1, representing the strength of the association in the current classroom teaching process. The association type of the corresponding semantic association is also read. For each edge, the edge identifier, association weight value, and association type are combined into an edge feature record. All edge feature records obtained after traversing all edges are collected to generate an edge feature record set.

[0048] The node feature record set is grouped and aggregated according to the knowledge point hierarchy category to obtain the category confidence statistics of each knowledge point hierarchy category. The edge feature set is grouped and aggregated according to the association relationship type to obtain the type weight statistics of each association relationship type. It needs to be specifically explained that all node feature records in the node feature record set are grouped according to knowledge point hierarchical categories, with node feature records belonging to the same knowledge point hierarchical category being grouped together. For each group, the entity confidence values ​​of all node feature records within that group are extracted, and the average and sum of these entity confidence values ​​are calculated as the category confidence statistic for that knowledge point hierarchical category. Specifically, the sum of all entity confidence values ​​within the group is divided by the number of nodes in the group to obtain the average confidence value for that knowledge category. This average confidence value is used to represent the overall importance of that knowledge point hierarchical category in the classroom teaching process. The calculated category confidence statistics for each knowledge category are then organized according to the knowledge point hierarchical category to form a category confidence statistics set.

[0049] All edge feature records in the edge feature record set are grouped according to their association relationship type, with edge feature records belonging to the same association relationship type grouped together. For each group, the association relationship weight values ​​of all edge feature records within that group are extracted, and the average and sum of these association relationship weight values ​​are calculated as the type weight statistics for that association relationship type. Specifically, the average weight value for that association relationship type is obtained by summing all association relationship weight values ​​within the group and dividing by the number of edges in the group. This average weight value is used to represent the overall association strength of that type of association in the classroom teaching process. The calculated type weight statistics for each association relationship type are then organized according to the association relationship type to form a type weight statistics set.

[0050] The category confidence statistics and type weight statistics are combined to obtain the graph feature set.

[0051] It is important to clarify that the set of category confidence statistics and the set of type weight statistics obtained in the above steps need to be merged. The merging operation involves organizing the two sets of statistics according to a unified data format to form a feature vector containing multiple feature dimensions. Each dimension of this feature vector corresponds to either a category confidence statistic for a knowledge point level or a type weight statistic for a relationship type. All the statistical values ​​of all dimensions are arranged according to a preset feature order to form a complete numerical sequence, which is the graph feature set. The graph feature set expresses the statistical characteristics of the temporal teaching knowledge subgraph at the node and edge levels in vector form, reflecting the importance of knowledge points at each level and the strength distribution of relationships between different types of knowledge in the classroom teaching process.

[0052] Based on the preset graph feature mapping rules corresponding to each teaching quality assessment dimension, the graph feature set is mapped to the corresponding teaching quality assessment dimension, generating the dimension feature value of each teaching quality assessment dimension. It is important to specify the acquisition of the pre-defined teaching quality assessment dimension template. This template contains multiple teaching quality assessment dimensions, each used to measure one aspect of classroom teaching quality. These dimensions include, but are not limited to: breadth of knowledge coverage (measuring the comprehensiveness of the knowledge points covered in the classroom); emphasis on key points (measuring whether core knowledge points are adequately emphasized in teaching); logical structure of knowledge (measuring the clarity of the logic between knowledge points according to their predecessor-successor and inclusion relationships); and sufficiency of teacher-student interaction (measuring the level of participation in question-and-answer interactions between teachers and students in the classroom). For each teaching quality assessment dimension, a set of graph feature mapping rules is pre-defined. These rules define how to select values ​​of one or more feature dimensions from the graph feature set to calculate the dimensional feature value of that teaching quality assessment dimension. For example, the mapping rule for the breadth of knowledge coverage dimension could be specified as selecting the sum or weighted sum of the category confidence statistics of all knowledge point hierarchical categories as the feature value of this dimension. The mapping rule for the emphasis on key points dimension could be specified as using the percentage of category confidence statistics of the core principle layer as the feature value of this dimension. Based on the mapping rules corresponding to each teaching quality assessment dimension, relevant statistical values ​​are extracted from the feature set of the graph, and dimensional feature values ​​for that dimension are generated according to a preset calculation method. The dimensional feature values ​​of all teaching quality assessment dimensions are then aggregated to generate the dimensional feature values ​​for each teaching quality assessment dimension.

[0053] Based on the pre-set set of teaching business constraint rules, the dimensional feature values ​​of each teaching quality evaluation dimension are constrained and verified. The dimensional feature values ​​that do not meet the constraint conditions are adjusted according to the corresponding constraint adjustment coefficient to obtain the constrained dimensional feature values. It is necessary to explain in detail the acquisition of the preset set of teaching business constraint rules. This set of teaching business constraint rules contains multiple teaching business constraint rules, each of which corresponds to a teaching quality evaluation dimension, specifying the range of values ​​or constraints that the feature values ​​of that dimension should satisfy.

[0054] For example, for the breadth of knowledge coverage dimension, business constraint rules can set a lower limit threshold for its dimensional feature values, as overly narrow knowledge coverage does not meet the basic requirements of classroom teaching. For each constraint rule, a corresponding constraint adjustment coefficient is also preset. This coefficient is used to adjust dimensional feature values ​​that do not meet the constraints to a reasonable range. For the dimensional feature values ​​of each teaching quality assessment dimension, the corresponding business constraint rules are used for verification. If the dimensional feature value meets the constraints, it remains unchanged. If the dimensional feature value does not meet the constraints, it is adjusted according to the corresponding constraint adjustment coefficient.

[0055] Based on the preset dimension weight coefficients of each teaching quality assessment dimension, the constrained dimension feature values ​​are weighted to obtain the comprehensive teaching quality score, and the intermediate teaching quality assessment results are summarized.

[0056] It's important to clarify that each dimension of teaching quality assessment corresponds to a weighting coefficient, which indicates the importance of that dimension in the overall teaching quality score. The weighting coefficients for different dimensions can be set according to specific teaching assessment needs. For example, the weighting coefficient for the breadth of knowledge coverage can be set to a higher value, indicating that the completeness of knowledge coverage is particularly important for teaching quality assessment; the weighting coefficient for the sufficiency of teacher-student interaction can be set to a lower value, indicating that while interaction needs to be considered, it is less important than the knowledge content itself. For each teaching quality assessment dimension, its corresponding constrained dimension feature value is multiplied by the dimension weighting coefficient to obtain the weighted feature value for that dimension. The weighted feature values ​​of all teaching quality assessment dimensions are summed to obtain the overall teaching quality score. This overall teaching quality score is a real number used to represent the overall quality level of the current classroom teaching process.

[0057] The calculated comprehensive teaching quality score serves as the core data for the intermediate teaching quality assessment. Simultaneously, the dimensional eigenvalues, constrained dimensional eigenvalues, and weighted eigenvalues ​​of each teaching quality assessment dimension are retained as supplementary information for the intermediate assessment. The comprehensive teaching quality score, along with the aforementioned supplementary information, is organized according to a pre-defined data format to form the intermediate teaching quality assessment result. This result includes the comprehensive classroom teaching quality score and detailed score information for each dimension, providing a data foundation for evidence tracing and visualization in subsequent steps.

[0058] S4: Map the intermediate evaluation results of teaching quality back to the time-series teaching knowledge subgraph to obtain a five-level traceable evidence chain, and generate a visual evaluation evidence graph to show the derivation path and basis of each evaluation conclusion.

[0059] In this embodiment of the invention, the constrained dimension feature values ​​of each teaching quality evaluation dimension are located from the time-series teaching knowledge subgraph, the associated target knowledge point entities and target association relationships are established, and a dimension-graph mapping record is created. It should be specifically noted that each constrained dimension feature value corresponds to a teaching quality evaluation dimension. This dimension is calculated from the statistical values ​​of one or more feature dimensions in the graph feature set. During the calculation of this dimension's feature value, the knowledge point entity or relationship corresponding to the graph feature dimension that contributes the most is recorded. Specifically, for the breadth of knowledge coverage dimension, its constrained dimension feature value may come from the sum of the category confidence statistics of each knowledge point level category, thus locating all knowledge point entities as the target knowledge point entities associated with this dimension. For the emphasis dimension, its constrained dimension feature value may come from the proportion of category confidence statistics of the core principle layer, thus locating knowledge point entities belonging to the core principle layer as the target knowledge point entities associated with this dimension. For the logicality of knowledge structure dimension, its constrained dimension feature value may come from the type weight statistics of predecessor-successor relationships, thus locating relationships with predecessor-successor relationships as the target relationships associated with this dimension. For each teaching quality assessment dimension, the dimension identifier, the corresponding constrained dimension feature value, and the list of target knowledge point entity identifiers and target relation identifiers are linked and recorded to form a mapping record for that dimension. All mapping records for all teaching quality assessment dimensions are then compiled to obtain a dimension-map mapping record.

[0060] Based on the comprehensive teaching quality score, constrained dimensional feature values, and dimensional-graph mapping records, a five-level traceable evidence chain is generated. It is important to clarify that evidence is extracted level by level, starting from the initial stage of the teaching process and following the data flow of the teaching assessment. The first level of evidence consists of the original teaching event records in the chronological teaching event set. These original teaching event records include all teaching event units collected during the classroom teaching process. Each teaching event unit contains an event timestamp, event type label, event participant identifier, and event text content. These original teaching event records are the most fundamental data source for all subsequent assessment conclusions.

[0061] Level 2 evidence consists of teaching trigger event records identified from the original teaching event logs. Teaching trigger event records are a summary list of newly added, disappeared, newly added, and disappeared trigger events identified from the chronological teaching event set. Each trigger event record includes the time of occurrence of the trigger event, the type of event triggered, or the participant identifier.

[0062] Level 3 evidence consists of records of entity confidence values ​​and relation weight values ​​for the temporal instructional knowledge graph, modified based on teaching-triggered events. These records contain the updated entity confidence value for each knowledge point entity and the updated relation weight value for each relation within the temporal instructional knowledge graph. These values ​​reflect the dynamic evolution of the knowledge graph during the classroom teaching process.

[0063] Level 4 evidence consists of a set of graph features extracted from the temporal instructional knowledge subgraph. This set includes category confidence statistics for each knowledge point level and type weight statistics for each type of association.

[0064] Level 5 evidence consists of records of intermediate teaching quality assessment results calculated based on the feature set of the graph. These records include the overall teaching quality score and the dimensional feature values ​​for each teaching quality assessment dimension. These five levels of evidence are organized in a progressive order from raw data to the final assessment results, forming a five-level traceable evidence chain.

[0065] The five-level traceable evidence chain specifically includes: a five-level progressive evidence sequence consisting of original teaching event records in the time-series teaching event set, teaching trigger event records identified from the original teaching event records, entity confidence values ​​and association weight values ​​of the time-series teaching knowledge subgraph modified according to the teaching trigger events, graph feature set records extracted from the time-series teaching knowledge subgraph, and intermediate teaching quality evaluation result records calculated based on the graph feature set. Each level of evidence in the five-level traceable evidence chain contains reference identifiers pointing to the previous level and the next level of evidence.

[0066] Based on the nodes and edges of the time-series teaching knowledge subgraph rendered by the five-level traceable evidence chain, a visual evaluation evidence graph is generated, which shows the derivation path and basis of each level of evidence in the five-level traceable evidence chain corresponding to the knowledge point entity or relationship.

[0067] Specifically, it is necessary to explain the acquisition of the hierarchical identifiers and content information of each level of evidence in the five-level traceable evidence chain. For each node in the time-series teaching knowledge subgraph, a hierarchical label is added to the node's visualization attributes based on the hierarchical position of the knowledge point entity corresponding to that node in the five-level traceable evidence chain. For example, if a knowledge point entity is recorded with an updated confidence value in the third-level evidence, the node is marked as a third-level evidence-related node; if a knowledge point entity is recorded as the source entity of the statistical value in the fourth-level evidence, the node is marked as a fourth-level evidence-related node. For each edge in the time-series teaching knowledge subgraph, a hierarchical label is added to the edge's visualization attributes based on the hierarchical position of the edge's corresponding association in the five-level traceable evidence chain. After adding the hierarchical labels, different visualization rendering attributes are configured for the nodes and edges according to the order of the evidence at each level in the five-level chain, including different colors, shapes, or marker symbols, so that nodes and edges of different evidence levels can be intuitively distinguished from the visualization graph. A graphics rendering engine is used to draw the time-series teaching knowledge subgraph with added hierarchical labels and rendering attributes into a visual interface, generating a visual evaluation evidence map.

[0068] like Figure 2The diagram shown is a functional module diagram of a knowledge graph-based intelligent evaluation system for teaching quality provided in an embodiment of the present invention.

[0069] The knowledge graph-based intelligent teaching quality evaluation system 100 described in this invention can be installed in an electronic device. Depending on the functions implemented, the knowledge graph-based intelligent teaching quality evaluation system 100 may include a graph construction and acquisition module 101, an event-driven update module 102, a feature mapping evaluation module 103, and an evidence visualization module 104. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0070] In this embodiment, the functions of each module / unit are as follows: The graph construction and acquisition module 101 is used to construct a subject-based basic knowledge graph containing knowledge point entities and inherent semantic relationships between knowledge point entities, collect the temporal teaching event stream during the classroom teaching process, and generate a temporal teaching event set. The event-driven update module 102 is used to identify teaching trigger events in the time-series teaching event set according to preset classroom time-driven update rules, modify the entity confidence and association weights within the subject basic knowledge graph according to the teaching trigger events, and generate a time-series teaching knowledge subgraph that evolves with the teaching process. The feature mapping evaluation module 103 is used to generate a graph feature set based on the entity features and relational features within the time-series teaching knowledge subgraph, map the graph feature set to the teaching quality evaluation dimension, and calculate the intermediate evaluation result of teaching quality in combination with teaching business constraints. The evidence visualization module 104 is used to map the intermediate evaluation results of teaching quality back to the time-series teaching knowledge subgraph to obtain a five-level traceable evidence chain and generate a visualized evaluation evidence map to show the derivation path and basis of each evaluation conclusion.

[0071] In the several embodiments provided by the present invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of modules is merely a logical functional division, and there may be other division methods in actual implementation.

[0072] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected to achieve the purpose of this embodiment according to actual needs.

[0073] In addition, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0074] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0075] The embodiments of this application can acquire and process relevant data based on an artificial intelligence technology. Artificial intelligence is a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A knowledge graph-based intelligent assessment method for teaching quality, characterized in that, The method includes: S1: Construct a subject-based basic knowledge graph containing knowledge point entities and inherent semantic relationships between knowledge point entities, collect the temporal teaching event flow in the classroom teaching process, and generate a temporal teaching event set; S2: Based on the preset classroom time-driven update rules, identify teaching trigger events in the time-series teaching event set, modify the confidence and association weights of entities within the subject basic knowledge graph according to the teaching trigger events, and generate a time-series teaching knowledge subgraph that evolves with the teaching process; S3: Based on the entity features and relational features within the temporal teaching knowledge subgraph, generate a graph feature set, map the graph feature set to the teaching quality assessment dimension, and calculate the intermediate teaching quality assessment results in conjunction with teaching business constraints. S4: Map the intermediate evaluation results of teaching quality back to the time-series teaching knowledge subgraph to obtain a five-level traceable evidence chain and generate a visual evaluation evidence graph to show the derivation path and basis of each evaluation conclusion.

2. The intelligent teaching quality assessment method based on knowledge graphs as described in claim 1, characterized in that, The construction of a subject-based knowledge graph that includes knowledge point entities and the inherent semantic relationships between knowledge point entities includes: Based on the standardized subject corpus of the target subject, the core concept system of the target subject is determined, the hierarchical structure of knowledge point entities is established, and the subject knowledge ontology is generated. Based on the subject knowledge ontology, knowledge point entities are extracted from the standardized subject corpus to obtain a set of knowledge point entities. Based on the subject knowledge ontology and the set of knowledge point entities, semantic relationships between knowledge point entities are extracted from the standardized subject corpus to obtain a set of entity relationships. Based on the set of knowledge point entities and the set of entity relationships, a subject-based knowledge graph is constructed, with knowledge point entities as nodes and semantic relationships as edges.

3. The intelligent teaching quality assessment method based on knowledge graphs as described in claim 2, characterized in that, The process of collecting sequential teaching events during classroom teaching generates a set of sequential teaching events, including: Multi-source event detection is performed on the original audio and video streams during classroom teaching to obtain multimodal teaching original event records, including teacher speech segments, student speech segments, blackboard content frames, and screen projection content frames; The teacher's speech segment and the student's speech segment are converted into text to obtain the teacher's speech text and the student's speech text. The blackboard content frame is subjected to optical character recognition to obtain the blackboard text. The screen projection content frame is parsed to obtain the projection text. The teacher's spoken text, student's spoken text, blackboard text, and projected text are aligned according to timestamps to generate a teaching text sequence. Semantic event boundary monitoring is performed on the teaching text sequence to obtain semantic turning points. The teaching text sequence is then segmented based on the semantic turning points to obtain teaching event units. Add event timestamps, event type tags, and event participant identifiers to the teaching event units, and combine the teaching event units in chronological order to obtain a set of time-series teaching events.

4. The intelligent teaching quality assessment method based on knowledge graphs as described in claim 3, characterized in that, The step of identifying teaching trigger events in the time-series teaching event set according to preset class time-driven update rules includes: Based on the chronological order of the event timestamps of each teaching event unit in the time-series teaching event set, the time-series teaching event set is divided into a continuous sequence of time windows. Based on the preset classroom time-driven update rules, the event type labels and event participant identifiers of the teaching event units within each time window are extracted from the time window sequence, and the event type distribution vector and participant distribution vector of the current window are generated. The event type distribution vector and participant distribution vector of the current window are compared with the historical event type distribution vector and historical participant distribution vector of the previous time window to obtain the change in event type and the change in participant distribution. Identify teaching-triggered events based on the changes in event type and participant distribution.

5. The intelligent teaching quality assessment method based on knowledge graphs as described in claim 4, characterized in that, The method of identifying teaching-triggered events based on changes in event type and changes in participant distribution includes: The changes in event types and participant distribution are analyzed to obtain the sets of event type addition identifiers, event type disappearance identifiers, participant addition identifiers, and participant disappearance identifiers. Based on the set of newly added event type identifiers and the set of event type disappearance identifiers, the first occurrence time and the last occurrence time of the event type within the current time window are marked as the event triggered by the addition of the event type and the event triggered by the disappearance of the event type, respectively. Based on the set of participant addition identifiers and the set of participant disappearance identifiers, the first time a participant participates and the last time a participant participates within the current time window are marked as the participant addition trigger event and the participant disappearance trigger event, respectively. Summarize all new trigger events of all types, trigger events for the disappearance of a type, trigger events for the addition of a participant, and trigger events for the disappearance of a participant to generate a list of teaching trigger events.

6. The intelligent teaching quality assessment method based on knowledge graphs as described in claim 5, characterized in that, The process of modifying the entity confidence and association weights within the subject-based knowledge graph based on teaching-triggered events to generate a temporal teaching knowledge subgraph that evolves with the teaching process includes: Based on the sequence of trigger event types in the list of teaching trigger events, determine the set of target knowledge point entities and the set of target association relationships corresponding to each teaching trigger event in the subject basic knowledge graph; Based on the trigger time sequence in the teaching trigger event list, perform a confidence increment operation on each target knowledge point entity in the target knowledge point entity set to generate an updated entity confidence value, and perform a weight increment operation on each relationship in the target relationship set to generate an updated relationship weight value. Update the subject-specific knowledge graph based on the new entity confidence value and the updated association weight value; Using the current time point of the classroom teaching process as the cutoff boundary, a time-series teaching knowledge subgraph is generated based on the knowledge point entities and relationships related to the current classroom time period in the updated subject basic knowledge graph.

7. The intelligent teaching quality assessment method based on knowledge graphs as described in claim 1, characterized in that, The step of generating a graph feature set based on the entity features and association features within the time-series teaching knowledge subgraph includes: Traverse all nodes of the time-series teaching knowledge subgraph, and generate a set of node feature records based on the entity confidence value of the knowledge point entity corresponding to each node and the knowledge point level category to which the entity belongs; Traverse all edges of the temporal teaching knowledge subgraph, and generate an edge feature set based on the association weight value and association type of the semantic association relationship corresponding to each edge; The node feature record set is grouped and aggregated according to the knowledge point hierarchy category to obtain the category confidence statistics of each knowledge point hierarchy category. The edge feature set is grouped and aggregated according to the association relationship type to obtain the type weight statistics of each association relationship type. The category confidence statistics and type weight statistics are combined to obtain the graph feature set.

8. The intelligent teaching quality assessment method based on knowledge graphs as described in claim 7, characterized in that, The process of mapping the graph feature set to the teaching quality assessment dimension, combined with teaching business constraints, and calculating the intermediate teaching quality assessment results includes: Based on the preset graph feature mapping rules corresponding to each teaching quality assessment dimension, the graph feature set is mapped to the corresponding teaching quality assessment dimension, generating the dimension feature value of each teaching quality assessment dimension. Based on the pre-set set of teaching business constraint rules, the dimensional feature values ​​of each teaching quality evaluation dimension are constrained and verified. The dimensional feature values ​​that do not meet the constraint conditions are adjusted according to the corresponding constraint adjustment coefficient to obtain the constrained dimensional feature values. Based on the preset dimension weight coefficients of each teaching quality assessment dimension, the constrained dimension feature values ​​are weighted to obtain the comprehensive teaching quality score, and the intermediate teaching quality assessment results are summarized.

9. The intelligent teaching quality assessment method based on knowledge graphs as described in claim 8, characterized in that, The process of mapping intermediate teaching quality assessment results back to a time-series teaching knowledge subgraph yields a five-level traceable evidence chain and generates a visualized assessment evidence graph, demonstrating the derivation path and basis for each assessment conclusion, including: From the temporal teaching knowledge subgraph, locate the constrained dimension feature values ​​of each teaching quality assessment dimension, the associated target knowledge point entities and target relationships, and establish a dimension-graph mapping record; Based on the comprehensive teaching quality score, constrained dimensional feature values, and dimensional-graph mapping records, a five-level traceable evidence chain is generated. The five-level traceable evidence chain specifically includes: a five-level progressive evidence sequence consisting of original teaching event records in the time-series teaching event set, teaching trigger event records identified from the original teaching event records, entity confidence values ​​and association weight values ​​of the time-series teaching knowledge subgraph modified according to the teaching trigger events, graph feature set records extracted from the time-series teaching knowledge subgraph, and intermediate teaching quality evaluation result records calculated based on the graph feature set; Based on the nodes and edges of the time-series teaching knowledge subgraph rendered by the five-level traceable evidence chain, a visual evaluation evidence graph is generated, which shows the derivation path and basis of each level of evidence in the five-level traceable evidence chain corresponding to the knowledge point entity or relationship.

10. A knowledge graph-based intelligent evaluation system for teaching quality, characterized in that, The system for implementing the knowledge graph-based intelligent assessment method for teaching quality as described in claim 1 includes: The graph construction and acquisition module is used to construct a subject-based basic knowledge graph containing knowledge point entities and the inherent semantic relationships between knowledge point entities, and to collect the temporal teaching event stream in the classroom teaching process to generate a temporal teaching event set. The event-driven update module is used to identify teaching trigger events in the time-series teaching event set according to the preset class time-driven update rules, modify the confidence and association weights of entities within the subject basic knowledge graph according to the teaching trigger events, and generate a time-series teaching knowledge subgraph that evolves with the teaching process. The feature mapping evaluation module is used to generate a graph feature set based on the entity features and relationship features within the time-series teaching knowledge subgraph, map the graph feature set to the teaching quality evaluation dimension, and calculate the intermediate evaluation results of teaching quality in combination with teaching business constraints. The evidence visualization module is used to map intermediate evaluation results of teaching quality back to the time-series teaching knowledge subgraph, obtain a five-level traceable evidence chain, and generate a visualized evaluation evidence map to show the derivation path and basis of each evaluation conclusion.

Images