Method for constructing psychological state monitoring model in dynamic interaction of learners

Abstract

The invention discloses a method for constructing a psychological state monitoring model in dynamic interaction of learners, which can monitor and analyze the dynamic psychological state of students in the learning process in real time. By combining emotion anomaly detection and psychological state modeling, emotion fluctuation of the student in the learning process can be accurately captured, and targeted intervention suggestions are provided, so that the learning effect and psychological health of the student are improved. Through a multi-level attention hierarchical structure, complex semantic information in student emotion expression can be fully mined, the weight is dynamically adjusted, and the accuracy of emotion analysis is improved. According to the method, an optimized long sequence modeling technology is utilized, long-time-span emotion changes in the learning process of students are captured, and the limitation that a traditional emotion analysis model cannot process long sequences is overcome.

Description

Technical Field

[0001] The present invention relates to the field of psychological state monitoring, and in particular to a method for constructing a psychological state monitoring model in dynamic interaction of learners. Background Art

[0002] With the rapid development of information technology, natural language processing (NLP) technology has made significant progress. The emergence of the Transformer architecture and the birth of pre-trained models such as BERT and GPT have brought breakthroughs in many NLP fields such as semantic analysis, sentiment recognition, question-answering systems, and machine translation, greatly improving the computer's ability to understand and process natural language, and providing strong support for solving various practical problems. At the same time, the field of education is also actively promoting informatization and intelligence. Online learning platforms are gradually becoming popular, and students' interactions with learning platforms are becoming more and more frequent, generating a large amount of text data, which provides a rich data foundation for using NLP technology to conduct research in the field of education.

[0003] In education management, monitoring students' mental state has always been a difficult problem. Students' emotional changes and mental health conditions are difficult to identify and intervene in a timely manner. Traditional mental health monitoring methods, such as questionnaires or face-to-face interviews, lack real-time and accuracy. With the widespread application of intelligent education and teaching assistance systems, multiple rounds of dialogue between students and the system generate a large amount of text data, providing new ways and possibilities for monitoring students' mental states. Therefore, how to use these text data and use NLP technology to effectively monitor students' mental states has become an important issue that needs to be solved in the field of education.

[0004] Since the emergence of the Transformer architecture, NLP technology has developed rapidly. In language environments such as English, related pre-trained models (such as BERT, RoBERTa, GPT) have been widely applied to a large number of practical problems, mostly focusing on the diversity and complexity of text, including sentiment classification and semantic understanding. However, there are still challenges in sentiment analysis and mental state monitoring of multi-turn dialogue data. The existing models' monitoring of sentiment is usually only effective in single-turn conversations, and the processing of long sequences of context and the recognition of emotional fluctuations still need to be further optimized. In particular, how to monitor students' mental health problems in real time based on emotional data is still a new research direction. For example, the paper "Bert: Pre-training of deep bidirectional transformers for language understanding" (authors: Kenton J D M W C, Toutanova L K; published in Proceedings of naacL-HLT. 2019, 1:2) introduced the relevant technologies of the BERT model in pre-training for language understanding. This model has an important position in the field of NLP, but it has deficiencies in aspects such as multi-turn dialogue sentiment analysis.

[0005] Compared with more mature technologies internationally, the application of NLP technology in the Chinese context in China started relatively late. Although models such as ERNIE and MacBERT have made progress in Chinese semantic understanding and generation, most research still focuses on short texts and specific task optimization, and the problems of semantic understanding and sentiment analysis of long sequences generated by multi-turn conversations have not been deeply solved. Domestic intelligent education platforms have begun to try to provide learning support based on multi-turn conversations, but the application of related research in sentiment analysis and mental state monitoring is still in its infancy. Most systems have not been able to fully utilize the interaction data between students and the platform for emotion recognition or mental state monitoring. At the same time, the research on mental health monitoring in China is not yet mature. The existing research mostly focuses on learning behavior analysis and performance evaluation, and there is less research on students' emotional fluctuations and mental health status. In terms of multi-turn dialogue data analysis and long-term mental state monitoring, no systematic solution has been formed. For example, the paper "ERNIE: Enhanced language representation with informative entities" (authors: Zhang Z, Han X, Liu Z, et al.; published in arXiv preprint arXiv:1905.07129, 2019) expounded on the technology of the ERNIE model in Chinese language representation, but it has limitations in specific application scenarios such as multi-turn dialogue long sequence sentiment analysis. Summary of the Invention

[0006] In view of the above problems, the present invention is proposed to provide a method for constructing a psychological state monitoring model in the dynamic interaction of learners to overcome the above problems or at least partially solve the above problems.

[0007] To solve the above technical problems, the embodiments of the present application disclose the following technical solutions:

[0008] A method for constructing a psychological state monitoring model in the dynamic interaction of learners, comprising:

[0009] The semantic understanding and attention mechanism module assigns attention weights at different text levels, enabling the monitoring system to focus on the key information of the learner. By means of weighted average or linear transformation, etc., it fuses the low-level detailed features and high-level abstract features, improving the monitoring system's understanding of complex semantics and the recognition ability of emotional states, enabling the model to better understand the emotional expressions and psychological state changes of students in multi-round conversations;

[0010] The long sequence modeling module adopts a sparse attention mechanism to selectively focus on key information. The sliding window attention slides on the long sequence to capture local context information, and the hierarchical attention analyzes the sequence at different levels, combining local and global information to construct a psychological state monitoring model that can effectively transmit emotional information in multi-round conversations and accurately identify changes in students' emotional states;

[0011] By adjusting the task objective and loss function, focusing on improving the model in emotional classification, and fine-tuning using a diverse training data set covering homework analysis data from different disciplines and conversation data of different types of learners, ensuring that the model can adapt to the homework analysis of different disciplines and provide accurate emotional classification and learning intervention suggestions for learners in different fields, realizing the optimization of the psychological state monitoring model.

[0012] Further, the semantic understanding and attention mechanism module assigns attention weights at different text levels, and the different text levels at least include words, phrases, sentences, and paragraphs.

[0013] Further, at the word level, pay attention to emotional keywords such as "anxiety" and "excitement" and learning-related vocabulary such as "homework" and "exam"; at the phrase level, focus on phrases expressing learning status and emotions such as "too much homework to finish" and "confident in the exam"; at the sentence level, grasp the emotions and learning conditions conveyed by complete sentences such as "I have been under a lot of learning pressure recently and feel on the verge of collapse"; at the paragraph level, combine the entire conversation paragraph to understand the emotional change trend of students in multi-round exchanges.

[0014] Further, at the word level, let the word vector of the word w i in the vocabulary V be x wi, calculate the attention weights through the training parameter matrix W w1 and W w2 Calculate the attention weights where b w1 and b w2 are bias terms, so as to focus on emotion and learning-related words and extract key information; at the phrase level, combine word vectors into a phrase vector x pi , use the parameter matrix W p1 and W p2 to calculate the attention weights where P is a set of phrases, focusing on phrases expressing learning status and emotion; at the sentence level, calculate the attention weight α si in a similar way through the sentence vector x si , grasp the sentence emotion and events; at the paragraph level, calculate the attention weights by integrating the information within the paragraph, fuse the features extracted at each level, and transform through a multi-layer perceptron to improve the model's ability to understand complex semantics and recognize emotions.

[0015] Furthermore, in the sparse attention mechanism, let the input sequence be X = [x 1 , x 2 ,..., x n , calculate the attention scores W q is the query matrix, d k is the scaling factor, and obtain the attention weights through the softmax function to select key information and reduce the computational complexity.

[0016] Furthermore, in the sliding window attention, let the window size be m, centered on the current round t, and focus on the information from round t - 2 / m to t + 2 / m. For the window sequence calculate the attention weight α w , to better understand the local context emotion intention.

[0017] Furthermore, for the hierarchical attention mechanism, first focus on the emotion semantics through α l within the local rounds, and comprehensively integrate the information of all rounds at the global level, and grasp the theme and emotion trend through α g , where at the global level effectively identify emotion information and accurately identify emotion changes.

[0018] Furthermore, by adjusting the task objectives and loss functions, focus on improving the model in sentiment classification, and fine-tune it using a diverse training dataset. The specific methods include: task-based fine-tuning based on the ERNIE model and fine-tuning dataset and task evaluation, optimizing its performance on specific tasks in the education field, leveraging the relevant features of the ERNIE-Multilevel model, and through sentiment analysis, learning disability identification, and academic dishonesty detection methods, the model needs to be able to accurately adapt to different task requirements.

[0019] Furthermore, adopt a multi-task learning strategy to integrate sentiment analysis, learning disability identification, and academic dishonesty detection tasks, sharing task objectives. In the learning disability identification task, students frequently make calculation errors, concept confusion, etc., and these features are associated with emotional features such as anxiety and depression in sentiment analysis, enabling the model to utilize the feature information of other tasks to enhance the sensitivity to students' emotional fluctuations. At the same time, use a diverse training dataset for fine-tuning, covering homework analysis data from different disciplines and dialogue data of different types of students, ensuring that the model can adapt to homework analysis in different disciplines and provide accurate sentiment classification and learning intervention suggestions for students in different fields. For students who have anxiety due to difficult problems in mathematics, the model can accurately identify and suggest that teachers provide targeted tutoring; for students who lack confidence in Chinese writing, the model suggests giving encouragement and writing skill guidance.

[0020] The beneficial effects of the above technical solutions provided by the embodiments of the present invention at least include:

[0021] A method for constructing a psychological state monitoring model in the dynamic interaction of learners disclosed by the present invention can improve the accuracy of semantic understanding and sentiment analysis: the multi-level semantic attention mechanism and the long sequence modeling optimization strategy enable the model to more accurately understand semantics and capture emotional changes in multi-round conversations, greatly improving the accuracy of sentiment analysis and effectively solving the deficiencies of existing models in this regard. It can also adapt to the requirements of education tasks. Through fine-tuning for education tasks and multi-task learning strategies, the model can better adapt to various tasks in the education field, real-time monitor mental health risks, and provide personalized intervention suggestions according to the risk situation. Compared with traditional mental health monitoring methods, it has higher real-time performance and accuracy.

[0022] The technical solutions of the present invention will be further described in detail below through the accompanying drawings and embodiments. Description of the Drawings

[0023] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation to the present invention. In the accompanying drawings:

[0024] Figure 1This is a flowchart of a method for constructing a psychological state monitoring model in dynamic interaction of learners in Embodiment 1 of the present invention;

[0025] Figure 2 This is a schematic structural diagram of an electronic device in Embodiment 2 of the present invention. Detailed implementation manners

[0026] Hereinafter, exemplary embodiments of the present disclosure will be described in more detail with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be completely conveyed to those skilled in the art.

[0027] To solve the problems existing in the prior art, an embodiment of the present invention provides a method for constructing a psychological state monitoring model in dynamic interaction of learners.

[0028] Embodiment 1

[0029] This embodiment discloses a method for constructing a psychological state monitoring model in dynamic interaction of learners, as Figure 1 , including:

[0030] The semantic understanding and attention mechanism module assigns attention weights at different text levels, enabling the monitoring system to focus on the key information of learners. By means of weighted average or linear transformation, etc., the low-level detailed features and high-level abstract features are fused to improve the monitoring system's understanding of complex semantics and the recognition ability of emotional states, enabling the model to better understand the emotional expressions and psychological state changes of students in multi-round conversations;

[0031] The long sequence modeling module adopts a sparse attention mechanism to selectively focus on key information. The sliding window attention slides on the long sequence to capture local context information, and the hierarchical attention analyzes the sequence at different levels, combining local and global information to construct a psychological state monitoring model that can effectively transmit emotional information in multi-round conversations and accurately identify changes in students' emotional states;

[0032] By adjusting the task objective and loss function, focusing on improving the model in emotional classification, and fine-tuning using a diverse training data set covering homework analysis data of different disciplines and conversation data of different types of learners, it is ensured that the model can adapt to the homework analysis of different disciplines and provide accurate emotional classification and learning intervention suggestions for learners in different fields, realizing the optimization of the psychological state monitoring model.

[0033] In this embodiment, the semantic understanding and attention mechanism module allocates attention weights at different text levels, and the different text levels include at least words, phrases, sentences, and paragraphs. Specifically, at the word level, focus on the emotional keywords "anxiety" and "excitement" and the vocabulary related to learning such as "homework" and "exam"; at the phrase level, focus on the phrases that express learning status and emotions such as "too much homework to finish" and "confident in the exam"; at the sentence level, grasp the emotions and learning status conveyed by complete sentences such as "I have been under a lot of pressure in my studies recently, and I feel like I am about to collapse"; at the paragraph level, combine the entire dialogue paragraph to understand the trend of students' emotional changes in multiple rounds of communication.

[0034] In this embodiment, at the word level, let word w in vocabulary V i The word vector is x wi , by training the parameter matrix W w1 and W w2 Calculating attention weights Among them, b w1 and b w2 is a bias term to focus on emotions and learn related vocabulary to extract key information; at the phrase level, the word vectors are combined into a phrase vector x pi , using the parameter matrix W p1 and W p2 Calculating attention weights Where P is a phrase set, focusing on phrases that express learning status and emotions; at the sentence level, the sentence vector x si The attention weight α is calculated in a similar way si , grasp the emotion and events of sentences; at the paragraph level, comprehensively calculate the attention weight of the information within the paragraph, fuse the features extracted at each level, and transform them through a multi-layer perceptron to enhance the model's ability to understand complex semantics and recognize emotions.

[0035] In this embodiment, in the sparse attention mechanism, let the input sequence be X = [x 1 , x 2 , ..., x n ], calculate the attention score W q is the query matrix, d k is the scaling factor, after the softmax function Get the attention weight, select key information, and reduce the amount of calculation.

[0036] In this embodiment, in terms of sliding window attention, the window size is set to m, with the current round t as the center, focusing on round information from t-2 / m to t+2 / m. For the window sequence Calculate the attention weight α w , better understand the emotional intention of the local context.

[0037] In this embodiment, for the hierarchical attention mechanism, first, within the local rounds, through α l focus on the emotional semantics, and at the global level, integrate the information of all rounds. Through α g grasp the theme and emotional trend. Among them, the global level effectively identify the emotional information and accurately identify the emotional changes.

[0038] In this embodiment, by adjusting the task objectives and loss functions, focus on improving the model in sentiment classification, and use diverse training datasets for fine-tuning. The specific methods include: performing task-based fine-tuning and fine-tuning dataset and task evaluation based on the ERNIE model to optimize its performance in specific tasks in the education field. Utilize the relevant characteristics of the ERNIE-Multilevel model. Through sentiment analysis, learning disability identification, and academic dishonesty detection methods, the model needs to be able to accurately adapt to different task requirements.

[0039] In this embodiment, adopt a multi-task learning strategy to integrate sentiment analysis, learning disability identification, and academic dishonesty detection tasks, and share the task objectives. In the learning disability identification task, students frequently make calculation errors, concept confusion, etc. These characteristics are associated with emotional characteristics such as anxiety and frustration in sentiment analysis, enabling the model to utilize the feature information of other tasks to enhance the sensitivity to students' emotional fluctuations. At the same time, use diverse training datasets for fine-tuning, covering homework analysis data from different disciplines and conversation data of different types of students, ensuring that the model can adapt to homework analysis in different disciplines and provide accurate sentiment classification and learning intervention suggestions for students in different fields. For students who have anxiety due to difficult problems in mathematics, the model can accurately identify and suggest that teachers provide targeted tutoring; for students who lack confidence in Chinese writing, the model suggests giving encouragement and writing skill guidance.

[0040] This embodiment discloses a method for constructing a psychological state monitoring model in the dynamic interaction of learners. By combining emotional anomaly detection and psychological state modeling, it can accurately capture the emotional fluctuations of students during the learning process and provide targeted intervention suggestions to improve students' learning effects and mental health.

[0041] This embodiment adopts a multi-level semantic attention mechanism. Through the multi-level attention hierarchy, it can fully mine the complex semantic information in students' emotional expressions and dynamically adjust the weights to improve the accuracy of sentiment analysis.

[0042] The method disclosed in this embodiment captures the emotional changes over a long time span during the learning process of students through long sequence modeling and by using the optimized long sequence modeling technology, overcoming the limitation that traditional sentiment analysis models cannot handle long sequences. By monitoring the emotional changes of students in real time and combining with the psychological state modeling module, personalized emotional intervention suggestions are provided for teachers to improve the emotional support and psychological counseling in the learning process. The ERNIE model is optimized and fine-tuned in multiple aspects for educational tasks to enhance its performance and adaptability in the educational field and provide personalized feedback and intervention.

[0043] Embodiment 2

[0044] Based on the same inventive concept, an embodiment of the present disclosure also provides an electronic device. Figure 2 It is a schematic structural diagram of an electronic device according to an embodiment of the present disclosure. As Figure 2 shown, an electronic device provided by an embodiment of the present disclosure includes: one or more processors 101, a memory 102, and one or more I / O interfaces 103. One or more programs are stored on the memory 102. When the one or more programs are executed by the one or more processors, the one or more processors implement any of the optimization methods in the above embodiments; one or more I / O interfaces 103 are connected between the processor and the memory and are configured to implement information interaction between the processor and the memory.

[0045] Among them, the processor 101 is a device with data processing capabilities, including but not limited to a central processing unit (CPU), etc.; the memory 102 is a device with data storage capabilities, including but not limited to a random access memory (RAM, more specifically such as SDRAM, DDR, etc.), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a flash memory (FLASH); the I / O interface (read / write interface) 103 is connected between the processor 101 and the memory 102 and can implement information interaction between the processor 101 and the memory 102, including but not limited to a data bus (Bus), etc.

[0046] In some embodiments, the processor 101, the memory 102, and the I / O interface 103 are connected to each other through a bus 104 and are further connected to other components of the computing device.

[0047] In some embodiments, the one or more processors 101 include a field programmable gate array.

[0048] According to an embodiment of the present disclosure, a computer-readable medium is also provided. A computer program is stored on the computer-readable medium. When the program is executed by a processor, the steps in any of the optimization methods in the above embodiments are implemented.

[0049] It should be understood that the specific order or hierarchy of steps in the disclosed process is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process can be rearranged without departing from the scope of the present disclosure. The appended method claims present the elements of the various steps in an exemplary order and are not intended to be limited to the specific order or hierarchy recited.

[0050] In the above detailed description, various features are combined in a single embodiment to simplify the present disclosure. This method of disclosure should not be interpreted as reflecting an intention that the embodiments of the claimed subject matter require more features than are expressly recited in each claim. Rather, as reflected in the appended claims, the invention lies in less than all of the features of a single disclosed embodiment. Accordingly, the appended claims are hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate preferred embodiment of the invention.

[0051] Those skilled in the art should also understand that the various illustrative logical blocks, modules, circuits, and algorithmic steps described in connection with the embodiments herein can be implemented as electronic hardware, computer software, or combinations thereof. To clearly illustrate the interchangeability of hardware and software, the various illustrative components, blocks, modules, circuits, and steps have been generally described in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as departing from the scope of the present disclosure.

[0052] The steps of a method or algorithm described in connection with the embodiments herein can be embodied directly in hardware, in a software module executed by a processor, or in a combination thereof. The software module can reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium well known in the art. An exemplary storage medium is coupled to the processor such that the processor can read information from, and write information to, the storage medium. Of course, the storage medium can also be integral to the processor. The processor and the storage medium can reside in an ASIC. The ASIC can reside in a user terminal. Of course, the processor and the storage medium can also exist as discrete components in a user terminal.

[0053] For software implementation, the techniques described in this application can be implemented by modules (e.g., procedures, functions, etc.) that perform the functions described in this application. These software codes can be stored in a memory unit and executed by a processor. The memory unit can be implemented within the processor or outside the processor. In the latter case, it is communicatively coupled to the processor via various means, which are well known in the art.

[0054] The above description includes examples of one or more embodiments. Of course, it is not possible to describe all possible combinations of components or methods for the purpose of describing the above embodiments, but those of ordinary skill in the art should recognize that the various embodiments can be further combined and arranged. Therefore, the embodiments described herein are intended to cover all such changes, modifications, and variations that fall within the scope of the appended claims. In addition, with respect to the term "comprising" as used in the specification or claims, this term is covered in a manner similar to the term "including" as interpreted when "including," is used as a transitional word in a claim. Further, any use of the term "or" in the specification or claims of a patent is to mean "non-exclusive or."

Claims

1. A method for constructing a psychological state monitoring model in learners' dynamic interaction, characterized in that: include: The semantic understanding and attention mechanism module allocates attention weights at different text levels, enabling the monitoring system to focus on key information of learners. It integrates low-level detail features and high-level abstract features through weighted average or linear transformation, improving the monitoring system's ability to understand complex semantics and recognize emotional states, enabling the model to better understand students' emotional expressions and changes in psychological states in multiple rounds of conversations. The long sequence modeling module uses a sparse attention mechanism to selectively focus on key information. The sliding window attention slides over the long sequence to capture local context information. The hierarchical attention analyzes the sequence from different levels and combines local and global information to build a psychological state monitoring model. The model effectively transmits emotional information in multiple rounds of dialogue and accurately identifies changes in students' emotional states. By adjusting the task objectives and loss functions, we focus on improving the model in sentiment classification and use a variety of training data sets for fine-tuning, covering homework analysis data from different disciplines and conversation data from different types of learners. This ensures that the model can adapt to homework analysis in different disciplines, provide accurate sentiment classification and learning intervention suggestions for learners in different fields, and optimize the mental state monitoring model.

2. A method for constructing a psychological state monitoring model in dynamic interaction of learners as claimed in claim 1, characterized in that: The semantic understanding and attention mechanism module allocates attention weights at different text levels, and the different text levels include at least words, phrases, sentences, and paragraphs.

3. The method for constructing a psychological state monitoring model in dynamic interaction of learners according to claim 1, characterized in that: At the word level, focus on the emotional keywords "anxiety" and "excitement" as well as the learning-related words "homework" and "exam"; At the phrase level, we focus on phrases that express learning status and emotions, such as "too much homework to finish" and "confident in the exam"; At the sentence level, grasp the emotions and learning status conveyed by complete sentences such as "I've been under a lot of academic pressure lately, and I feel like I'm about to collapse"; at the paragraph level, combine the entire dialogue paragraph to understand the trend of students' emotional changes in multiple rounds of communication.

4. The method for constructing a psychological state monitoring model in dynamic interaction of learners according to claim 1, characterized in that: At the word level, let word w in vocabulary V i The word vector is x wi , by training the parameter matrix W w1 and W w2 Calculating attention weights Among them, b w1 and b w2 is a bias term to focus on emotions and learn related vocabulary to extract key information; at the phrase level, the word vectors are combined into a phrase vector x pi , using the parameter matrix W p1 and W p2 Calculating attention weights Where P is a phrase set, focusing on phrases that express learning status and emotions; at the sentence level, the sentence vector x si The attention weight α is calculated in a similar way si , grasp the emotion and events of sentences; at the paragraph level, comprehensively calculate the attention weight of the information within the paragraph, fuse the features extracted at each level, and transform them through a multi-layer perceptron to enhance the model's ability to understand complex semantics and recognize emotions.

5. The method for constructing a psychological state monitoring model in dynamic interaction of learners according to claim 1, characterized in that: In the sparse attention mechanism, let the input sequence be X = [x1, x2, ..., x n ], calculate the attention score W q is the query matrix, d k is the scaling factor, after the softmax function Get the attention weight, select key information, and reduce the amount of calculation.

6. The method for constructing a psychological state monitoring model in dynamic interaction of learners according to claim 1, characterized in that: In terms of sliding window attention, let the window size be m, center on the current round t, and focus on round information from t-2 / m to t+2 / m. For the window sequence Calculate the attention weight α w , better understand the emotional intention of the local context.

7. The method for constructing a psychological state monitoring model in dynamic interaction of learners according to claim 1, characterized in that: For the layered attention mechanism, first pass α in the local round l Focusing on sentiment semantics, we integrate all round information at the global level and g Grasp the theme and emotional direction, among which, Global level Effectively identify emotional information and accurately identify emotional changes.

8. The method for constructing a psychological state monitoring model in dynamic interaction of learners according to claim 1, characterized in that: By adjusting the task objectives and loss functions, we focus on improving the model in sentiment classification and use a variety of training data sets for fine-tuning. Specific methods include: performing task-based fine-tuning and fine-tuning data sets and task evaluation based on the ERNIE model to optimize its specific task performance in the field of education, and utilizing the relevant characteristics of the ERNIE-Multilevel model. Through sentiment analysis, learning disability identification, and academic dishonesty detection methods, the model needs to be able to accurately adapt to different task requirements.

9. The method for constructing a psychological state monitoring model in dynamic interaction of learners according to claim 1, characterized in that: A multi-task learning strategy is adopted to integrate sentiment analysis, learning disability identification, and academic dishonesty detection tasks, and share task objectives. In the learning disability identification task, students frequently make calculation errors and conceptual confusions. These characteristics are associated with emotional characteristics such as anxiety and depression in sentiment analysis, so that the model can use the feature information of other tasks to improve its sensitivity to students' emotional fluctuations. At the same time, a variety of training data sets are used for fine-tuning, covering homework analysis data of different subjects and conversation data of different types of students, to ensure that the model can adapt to homework analysis of different subjects and provide accurate sentiment classification and learning intervention suggestions for students in different fields. For students who have anxiety due to difficult problems in mathematics, the model can accurately identify them and suggest that teachers provide targeted tutoring. For students who lack confidence in Chinese writing, the model recommends encouragement and writing skills guidance.

10. An electronic device comprising: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the mental state monitoring model construction method.

Images