A method for identifying teaching behavior events in asynchronous delivery classrooms based on teaching reflection
By constructing an asynchronous specialized classroom teaching behavior event recognition method, using preprocessing and feature fusion technology, the problem of teaching behavior event recognition in remote areas is solved, and the automated analysis of teaching reflection text is realized, which is suitable for teaching data collection and analysis in remote areas, reducing the number of teaching data Cost and hardware requirements.
Patent Information
- Application Number
- CN202411304164.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-19
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-09-19
AI Technical Summary
The teaching process of asynchronous specialized classrooms in remote areas is unclear, and there is a lack of effective teaching behavior event recognition methods, making it difficult to realize automated analysis of teaching reflection texts.
Construct an asynchronous specialized classroom teaching behavior event recognition method based on teaching reflection text. By preprocessing teaching reflection text, extract word embedding features and context features, and perform multiple splicing and fusion, combining features with the maximum pooling layer to extract features, realize teaching behavior events Automatic identification.
Effectively identifying teaching behavior events reduces labor costs, improves the degree of automation of teaching reflection text analysis, and is suitable for teaching data collection and analysis in remote areas.
Smart Images

Figure CN118820911B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a method for identifying teaching behavior events in an asynchronous special delivery classroom based on teaching reflection, and belongs to the field of educational information technology. Background Art
[0002] Delivering distance education video teaching materials to remote areas with limited educational conditions to conduct asynchronous special delivery classes is one of the effective ways to promote educational equity. How teachers in remote areas carry out teaching with the help of video teaching materials is a key factor in determining the quality of teaching. Teaching equipment in remote areas is limited, and it is difficult to record and identify the process through video recording. Teaching behavior events are the components of classroom teaching and the basic units for teachers to carry out teaching reflection. Teachers' post-class teaching reflections record teachers' memories of the teaching process, their thoughts on teaching problems, and their expressions of teaching concepts. Therefore, teaching reflection texts can become another channel for remote areas to understand the teaching process through the identification of teaching behavior events. In addition, through the analysis of teaching behavior events, key behavior events in teachers' professional development can be defined, and the depth and breadth of teachers' teaching reflections can be analyzed.
[0003] It can be seen that the recognition of teaching behavior events in asynchronous special delivery classrooms based on teaching reflection texts can not only help teachers, managers, and researchers to timely understand the teaching implementation process in areas with limited teaching conditions, but also provide a convenient method for collecting continuous data for defining key teaching events in the process of teacher professional development, and provide a reference for automated analysis of teaching reflections. Therefore, the present invention proposes a method for identifying teaching behavior events in asynchronous special delivery classrooms based on teaching reflection texts. The method proposes a preprocessing method for teaching reflection texts and an encoding of teaching behavior events, thereby constructing an asynchronous special delivery classroom teaching behavior event recognition data set; based on the asynchronous special delivery classroom teaching behavior event recognition data set, a teaching behavior event recognition model is constructed that multiple splices and fuses the recognition text context features and word embedding features, thereby realizing automatic recognition of teaching behavior events in asynchronous special delivery classroom teaching reflection texts. Summary of the invention
[0004] The technical problem to be solved by the present invention is to provide a method for identifying teaching behavior events in an asynchronous special delivery classroom based on teaching reflection, which is used to solve the problem of unclear teaching process in asynchronous special delivery classrooms in remote areas, and can also provide a preliminary methodological basis for the automated analysis of teaching reflection texts.
[0005] The technical solution of the present invention is: a method for identifying teaching behavior events in an asynchronous special delivery classroom based on teaching reflection, constructing an asynchronous special delivery classroom teaching behavior event recognition data set based on teaching reflection text, and proposing a teaching behavior event recognition model. The model extracts context features and word embedding features of the teaching reflection text and performs multiple splicing and fusion. The fused features are then passed through a maximum pooling layer to extract salient features, thereby performing teaching behavior event label prediction and realizing automatic recognition of teaching behavior events in teaching reflection texts.
[0006] The specific steps of the method are:
[0007] Step 1: Obtain the text data of teachers’ teaching reflections in asynchronous delivery classrooms.
[0008] Step 2: Based on the teaching reflection text data, construct an asynchronous delivery classroom teaching behavior event recognition dataset.
[0009] The Step 2 is specifically as follows:
[0010] Step 2.1: Pre-process the emoji emoticons and punctuation marks in the text data of the teacher's teaching reflection in the asynchronous delivery classroom, specifically:
[0011] Step 2.1.1: Convert the Excel file of the asynchronous delivery classroom teacher's teaching reflection text into a txt file (UTF-8 encoding), convert the emoji emoticons in the teaching reflection text into regular punctuation marks and remove the regular punctuation marks generated by the emoji emoticon conversion.
[0012] Step 2.1.2: Add punctuation marks to the long text without punctuation marks and correct the text data.
[0013] Step 2.1.3: Select punctuation marks according to the required granularity to segment the text data. Suppose the sample set obtained after preprocessing and segmenting the teaching reflection text data written by teachers after class is , S is the sentence divided by punctuation marks in the teaching reflection text, and N is the number of samples.
[0014] Step 2.2: Based on basic teaching links such as "teacher presentation", "student practice", "evaluation" and "regulatory feedback", refine the teaching behavior events under the teaching links and construct the asynchronous special delivery classroom teaching behavior event coding.
[0015] Step 2.3: Based on the asynchronous delivery classroom teaching behavior event coding constructed in Step 2.2, annotate the text to be annotated obtained in Step 2.1.3 to obtain the asynchronous delivery classroom teaching behavior event recognition dataset , text data sample , where X represents the words that make up the sentence sample S, and m represents the number of words in the text data sample. It is divided into training set, test set and validation set according to 6:2:2 as the training material for the teaching behavior event recognition model.
[0016] The Step 3 is specifically as follows:
[0017] Step 3: Based on the data set constructed in Step 2.3, build a teaching behavior event recognition model based on deep learning technology.
[0018] Step 3.1: Use the asynchronous classroom teaching behavior event recognition dataset constructed in Step 2.3 as the input of the pre-trained Bert model, and extract the word embedding features of the recognition text S through the pre-trained Bert model. , where S e is the word embedding feature extracted from sample S, X e The word embedding features extracted for word X, m represents the number of words in the text data sample.
[0019] Step 3.2: embed the word extracted in Step 3.1 into feature S e As the input of the long short-term memory network BiLSTM, the context features of the recognized text are extracted through BiLSTM. The extracted context features are expressed as , t represents the time step.
[0020] Step 3.3: Multiple concatenation and fusion of the context features extracted by BiLSTM in Step 3.2 and the word embedding features extracted by Bert in Step 3.1 to obtain temporal semantic enhancement features, specifically:
[0021] The context feature and the word embedding feature are first concatenated and fused to obtain a temporal semantic fusion feature. ;
[0022] Then combine the temporal semantic fusion feature with the word embedding feature S e Secondary splicing and fusion to obtain temporal semantic enhancement features ;
[0023] Step 3.4: Take the text temporal semantic enhancement features obtained in Step 3.3 as the basis for teaching behavior event recognition. First, perform Max pooling operation to obtain the pooled features after dimensionality reduction:
[0024]
[0025] The pooled features are used as the input of the fully connected layer to obtain the fully connected layer features, and the output result is:
[0026] ,
[0027] In the formula, is the weight matrix, is the bias vector;
[0028] Finally, the output of the fully connected layer The feature vector is input into the softmax layer to obtain the predicted teaching behavior event code , complete the recognition of teaching behavior events of recognized text.
[0029] In the Step 2.1, the emoji emoticons in the teaching reflection text are processed by converting the file format and removing punctuation marks, so that the recognized text is composed of regular text and punctuation marks, reducing the influence of emoji emoticons on text feature extraction and semantic recognition results, which is beneficial to the recognition model to extract the semantic features of the recognized text.
[0030] At the same time, the text data is corrected by adding punctuation marks to long texts without punctuation marks;
[0031] Finally, the teaching reflection text is segmented by punctuation marks to control the granularity of the recognized teaching reflection text.
[0032] In the Step 3.3, a method of multiple splicing and fusion of text context features and word embedding representation features is proposed, specifically:
[0033] The context feature and the word embedding feature are first concatenated and fused to obtain a temporal semantic fusion feature. ;
[0034] Then combine the temporal semantic fusion feature with the word embedding feature S e Secondary splicing and fusion to obtain temporal semantic enhancement features .
[0035] This fusion method is the core of the teaching behavior event recognition model. It extracts the contextual features and word embedding features of the text through BiLSTM, BERT, etc., and then processes the obtained semantic features in a fusion way to obtain enhanced text representation, thereby improving the accuracy of text classification.
[0036] In the Step 3.4, the teaching behavior event to which the text belongs is predicted based on the fusion features in Step 3.3, and the local significant features of the text are extracted through feature maximum pooling to remove redundant information and improve the generalization ability of the model in the recognition task.
[0037] The present invention designs a method for identifying teaching behavior events in an asynchronous special delivery classroom based on teaching reflection text, and applies text classification technology in natural language processing to the task of identifying teaching behavior events. The method designs a preprocessing method for emoji emoticons and punctuation marks in teaching reflection text; constructs an asynchronous special delivery classroom teaching behavior event encoding; takes information technology courses as an example, constructs an asynchronous special delivery classroom teaching behavior event recognition data set; extracts text word embedding features and context features, and uses multiple splicing and fusion of features to optimize the effect of the model in the task of identifying teaching behavior events.
[0038] Compared with general areas, it is more difficult to collect teaching data in remote areas, and it is more difficult to conduct continuous quantitative analysis of teaching conditions. Since teaching reflection texts contain teachers' recollections of the teaching process, thoughts on teaching problems, and expressions of teaching concepts, they provide a data basis for understanding the teaching process in remote areas. The invention designs an automated analysis method for teaching reflection texts, from the teaching reflection text preprocessing method, the teaching behavior event coding design, the construction of a teaching reflection text data set for asynchronous special delivery classroom teaching behavior event recognition to the construction of a classification model in the field of natural language processing. This has important reference significance for realizing the analysis of the teaching process in remote areas based on teaching reflection texts.
[0039] The beneficial effects of the present invention are: it can effectively identify teaching behavior events in the teaching reflection text of the asynchronous special delivery classroom of the information technology course. Compared with the existing teaching behavior event recognition method, the software and hardware requirements for data acquisition and storage are low, and it is more suitable for rural areas. Compared with the existing teaching reflection text analysis method based on manual labor, the present invention can reduce the labor cost of the teaching reflection text encoding link in the existing analysis method and realize automation. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] Figure 1 is a flow chart of the present invention;
[0041] Figure 2 It is the preprocessing process of the multi-emoji teaching reflection text of the present invention;
[0042] Figure 3 It is the asynchronous special delivery classroom teaching behavior event coding of the present invention;
[0043] Figure 4 It is an example of the information technology course asynchronous special delivery classroom teaching behavior event recognition data set of the present invention;
[0044] Figure 5 The information technology course asynchronous special delivery classroom teaching behavior event recognition data set construction of the present invention;
[0045] Figure 6 It is a structural diagram of the teaching behavior event recognition model of the present invention. DETAILED DESCRIPTION
[0046] The present invention will be further described below in conjunction with the accompanying drawings and specific implementation methods.
[0047] like Figure 1 As shown in FIG. 1 , a method for identifying teaching behavior events in an asynchronous delivery classroom based on teaching reflection is provided, and the specific steps are as follows:
[0048] Step 1: Obtain the teaching reflection text data of teachers in asynchronous special delivery classrooms, and export the teaching reflection text data of teachers in asynchronous special delivery classrooms from the online platform;
[0049] Step 2: Construct an asynchronous classroom teaching behavior event recognition dataset based on teaching reflection text;
[0050] Step 2.1: Pre-process the emoji emoticons and punctuation marks in the text data of the teacher's teaching reflection in the asynchronous delivery classroom:
[0051] Step 2.1.1: Convert the Excel file of the teacher's teaching reflection text in the asynchronous delivery classroom obtained in Step 1 into a txt file, set the encoding method to UTF-8, convert the emoji emoticons in the teaching reflection text into regular punctuation marks and remove them, so that the recognized text consists of regular text and punctuation marks, which facilitates the recognition model to extract the semantic features of the recognized text.
[0052] Step 2.1.2: Use the Paddlehub package in the PaddlePaddle deep learning framework to add punctuation to long texts without punctuation, correct the data samples, and make the punctuation in the teaching reflection text conform to the usage habits of Chinese punctuation.
[0053] Step 2.1.3: Segment the obtained teaching reflection text according to punctuation marks to obtain a sample set of teaching reflection texts to be annotated. , S is the sentence segmented by punctuation marks in the teaching reflection text, and N is the number of samples. When segmenting sentences, the choice of punctuation marks can control the granularity of the recognized teaching reflection text.
[0054] Step 2.2: Based on the basic teaching links of "teacher presentation", "student practice", "evaluation" and "regulation and feedback", construct the asynchronous special delivery classroom teaching behavior event coding.
[0055] The "teacher presentation" link is refined. This link mainly introduces teaching by creating teaching situations, creating questions, and reviewing old knowledge. New knowledge is taught by playing educational video teaching. The new knowledge in the video teaching material is taught again by demonstrating the operation of new knowledge, conducting questions and answers, and organizing discussions, so as to ensure that students have a clear understanding of the teaching content of this lesson. Therefore, the present invention subdivides "teacher presentation" into three teaching behavior event types: "teaching introduction", "educational video teaching material learning", and "teaching assistant teaching".
[0056] The "regulation and feedback" link is refined. The regulation and feedback link based on teacher interaction mainly includes discovering teaching problems through teachers patrolling the classroom and students raising their hands to ask questions, solving teaching problems through guiding students to review educational video materials, mutual help among students and guidance from teaching assistants, displaying and communicating through student works, mutual evaluation among students and students expressing their thoughts and opinions, and evaluating students' learning and mastery through classroom summaries by teachers and students combing through teaching content. Therefore, the study subdivided the regulation and feedback link into four teaching behavior events: "discovering problems", "solving problems", "display and communication" and "classroom summary".
[0057] In addition to the teaching process, combined with the actual implementation of the asynchronous special delivery classroom of the information technology course, the present invention supplements the teaching behavior events, adding teaching behavior events such as "teaching preparation", "classroom management", "review and consolidation", "games and stories" and "homework".
[0058] In order to distinguish the texts that do not contain teaching behavior event descriptions, the code "No teaching behavior event" was added. Finally, the coding system contains 15 teaching behavior event codes, as shown in the attached figure. Figure 2 shown.
[0059] Step 2.3: Taking the information technology course as an example, based on the asynchronous special delivery classroom teaching behavior event coding constructed in Step 2.2, annotate the text to be annotated obtained in Step 2.1.3 to obtain the information technology course asynchronous special delivery classroom teaching behavior event recognition dataset , text data sample , where X represents the words that make up the sentence sample S, and m represents the number of words in the text data sample.
[0060] Constructed dataset It contains 33,292 data items, about 456,000 characters. The dataset examples and dataset construction are shown in the attached figure. Figure 3 , 4 shown.
[0061] The dataset is divided into training set, test set and validation set according to the ratio of 6:2:2, which serves as the training material for the teaching behavior event recognition model.
[0062] Step 3: Based on the asynchronous classroom teaching behavior event recognition dataset constructed in Step 2.3, a teaching behavior event recognition model based on deep learning technology is proposed. The structure is shown in the figure Figure 5 .
[0063] Step 3.1: Use the asynchronous special delivery classroom teaching behavior event recognition dataset divided in Step 2.3 as the input of the pre-trained Bert model, and extract the word embedding features of the recognition text S through the pre-trained Bert model , where S e is the word embedding feature extracted from sample S, X e The word embedding features extracted for word X include symbol embedding, sentence embedding, and position embedding features. m represents the number of words in the recognized text. The length of each sentence processed is set to 32, that is, m is 32, and the word embedding feature dimension composed of the three features is 768.
[0064] Step 3.2: Embed the words extracted in Step 3.1 The features are used as the input of the long short-term memory network BiLSTM, and the context features of the recognized text are extracted using BiLSTM. The extracted context features are expressed as , t represents the time step, and the feature dimension is 1536.
[0065] Step 3.3: The context features extracted by BiLSTM in Step 3.2 and the word embedding features extracted by Bert in Step 3.1 are fused multiple times, specifically:
[0066] The context feature and the word embedding feature are first concatenated and fused to obtain a temporal semantic fusion feature. , the feature dimension size is 2304;
[0067] Then combine the temporal semantic fusion feature with the word embedding feature S e Secondary splicing and fusion to obtain temporal semantic enhancement features , the feature dimension is 3072;
[0068] This fusion method is the core of the teaching behavior event recognition model. It extracts the contextual features and word embedding features of the text through BiLSTM, BERT, etc., and then processes the obtained semantic features using a splicing and fusion method to obtain enhanced text representation, thereby improving the accuracy of text classification.
[0069] Step 3.4: Take the comprehensive features of the teaching reflection text obtained in Step 3.3 as the basis for teaching behavior event recognition. First, perform the Max pooling operation to obtain the pooled features after dimensionality reduction:
[0070]
[0071] The pooled features are used as the input of the fully connected layer to obtain the fully connected layer features, and the output result is:
[0072]
[0073] In the formula, is the weight matrix, is the bias vector;
[0074] Finally, the output of the fully connected layer The feature vector is input into the softmax layer to obtain the predicted teaching behavior event code , complete the recognition of teaching behavior events of recognized text.
[0075] The experiment was carried out on a server with a GPU model of NVIDIA GeForce RTX 3090 and a memory of 4G. The performance results of each model are shown in Table 1. In summary, the model of multiple fusion of BiLSTM and Bert features proposed in the present invention shows certain advantages in the task of teaching behavior event recognition of Chinese teaching reflection texts.
[0076] Table 1 Comparative experimental results (Macro average)
[0077]
[0078] The specific implementation modes of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above implementation modes, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.
Claims
1. A method for identifying teaching behavior events in asynchronous classroom based on teaching reflection, characterized in that: The method comprises: Step 1: Obtain the teaching reflection text data of teachers in asynchronous delivery classroom; Step 2: Based on the teaching reflection text data, construct an asynchronous delivery classroom teaching behavior event recognition dataset; Step 3: Based on the asynchronous special delivery classroom teaching behavior event recognition dataset, a teaching behavior event recognition model based on deep learning technology is constructed to complete the recognition; The Step 3 is specifically as follows: Step 3.1: Use the Bert pre-training model to identify the text sample S in the asynchronous special delivery classroom teaching behavior event recognition dataset, extract the word embedding features of the text sample S to represent the semantic information, and obtain: ; Among them, S e To identify the word embedding features extracted from the text S, X e The word embedding features extracted for word X, m represents the number of words in the recognized text; Step 3.2: Based on word embedding feature S e , the context features of the recognized text are extracted by Bi-LSTM to represent the temporal information. The extracted context features are expressed as: , t represents the time step; Step 3.3: Multiple concatenation and fusion of the context features and the word embedding features to obtain temporal semantic enhancement features; Step 3.4: Perform Max pooling operation on the temporal semantic enhancement features to obtain the pooled features after dimensionality reduction: ; The pooled features are used as the input of the fully connected layer to obtain the fully connected layer features, and the output result is: ; In the formula, is the weight matrix, is the bias vector; Finally, the output of the fully connected layer The feature vector is input into the softmax layer to obtain the predicted teaching behavior event code , complete the recognition of teaching behavior events of recognized text.
2. The method for identifying teaching behavior events in asynchronous classroom based on teaching reflection according to claim 1 is characterized in that: The Step 2 is specifically as follows: Step 2.1: Preprocess the emoji emoticons and punctuation marks in the text data of teachers’ teaching reflections in asynchronous special delivery classrooms. Suppose the sample set obtained after preprocessing the text data of teachers’ teaching reflections written after class is There are N samples in , each sample is a sentence divided by punctuation marks in the teaching reflection text; Step2.2: Construct the teaching behavior event coding of asynchronous special delivery classroom; Step 2.3: Based on the asynchronous special delivery classroom teaching behavior event coding, annotate the asynchronous special delivery classroom teaching reflection text, and construct the asynchronous special delivery classroom teaching behavior event recognition data set. The annotated sample set is , text sample , X represents the words that make up the sentence sample S, and m represents the number of words in the text data sample.
3. The method for identifying teaching behavior events in asynchronous classroom based on teaching reflection according to claim 2 is characterized by: In the Step 2.1, the emoji emoticons in the teaching reflection text are processed by converting the file format and removing the punctuation marks, so that the recognized text consists of regular text and punctuation marks; At the same time, the text data is corrected by adding punctuation marks to long texts without punctuation marks; Finally, the teaching reflection text is segmented by punctuation marks to control the granularity of the recognized teaching reflection text.
4. The method for identifying teaching behavior events in asynchronous classroom based on teaching reflection according to claim 1 is characterized in that: In Step 3.3, the text context features and word embedding features are concatenated and fused multiple times, specifically: The context feature and the word embedding feature are first concatenated and fused to obtain a temporal semantic fusion feature. ; Then combine the temporal semantic fusion feature with the word embedding feature S e Secondary splicing and fusion to obtain temporal semantic enhancement features .
5. The method for identifying teaching behavior events in asynchronous classroom based on teaching reflection according to claim 1 is characterized by: In Step 3.4, the teaching behavior event to which the text belongs is predicted through temporal semantic enhancement features, and the local significant features of the recognition text are extracted through feature maximum pooling to remove redundant information and improve the generalization ability of the model in recognition tasks.
