Vocational education personalized learning assistant construction method based on virtual digital human

Abstract

The invention discloses a vocational education personalized learning assistant construction method based on virtual digital humans, which is characterized by comprising the following steps: inputting vocational education-related user information into a large language model for processing, outputting a course name, a degree value of each index and a user intention, and if the course name does not exist in a preset course library list, judging whether the course name does not exist in the preset course library list or not; adding the course into a course library list if the course exists, and updating the course library list if the course exists; forming a text by the output course name, the degree value of each index and the user intention, inputting the text into a Text2SQL model, and retrieving in a database to obtain a recommended learning resource list; and finally, screening the learning resource list and the updated course library list by adopting a decision tree model to obtain a learning resource list of which the correlation coefficient is greater than 0.7, and inputting the learning resource list and user information into a cross encoder for processing to obtain reordered learning resource list recommendation. The method has the advantage that dynamic and personalized learning path recommendation and real-time feedback can be realized.

Description

Technical Field

[0001] The present invention relates to a personalized learning assistant for vocational education, and more particularly to a method for constructing a personalized learning assistant for vocational education based on virtual digital humans. Background Art

[0002] With the continuous development of information technology, especially the rapid progress of artificial intelligence, virtual reality (VR) and natural language processing (NLP) technologies, the teaching mode in the field of vocational education is gradually undergoing profound changes. The traditional vocational education mode mostly relies on face-to-face teaching and classroom lectures. With the penetration of technology, the hybrid teaching mode combining online and offline has become a new trend. In this process, virtual digital human technology has been more and more widely used and has become an important auxiliary tool in the teaching and learning process.

[0003] As an intelligent virtual assistant, a virtual digital human can interact with students through natural language communication. Its characteristics are that it can simulate the voice, facial expressions, body movements, etc. of real people, enabling students to obtain an immersive learning experience. In vocational education, virtual digital humans can not only act as tools for imparting basic knowledge, but also interact and answer questions in some simple scenarios. For example, virtual digital humans can provide a brief review of course content, answer questions, and even guide students to complete some operation tasks, greatly improving learning efficiency and participation. At present, virtual digital human technology has been applied in many fields, such as education, healthcare and customer service. In the field of education, virtual digital humans are widely used in the auxiliary of basic teaching and the link of answering students' questions. Their main role is to help students understand some simple concepts or complete some basic operations. Especially in large-scale online education, virtual digital humans can effectively replace traditional manual explanations and tutoring, reducing educational costs and improving the utilization rate of educational resources.

[0004] However, despite the great potential shown by virtual digital humans in the field of education, especially the significant progress made in efficiently transmitting knowledge and solving basic problems, the application of existing technologies is still far from the ideal state of personalized learning. In vocational education, as intelligent learning assistants, virtual digital humans are more at the level of information transmission and have not been able to fully consider the individual differences of learners and the specific needs of vocational skills. In recent years, the rise of large language models (LLMs) has brought new ideas to the development of virtual digital humans. LLM technologies represented by GPT have demonstrated powerful semantic understanding and reasoning capabilities in the field of natural language processing. LLMs have been widely applied to conversational artificial intelligence agents, capable of generating highly natural human-like texts, bringing a more intuitive and intelligent experience to user interactions. This technological progress provides the possibility of integrating LLMs into recommendation systems, significantly enhancing the user experience by capturing user preferences in real time, understanding the context, and generating comprehensive personalized recommendations.

[0005] Although the introduction of LLM technology has provided new possibilities for recommendation systems, existing technologies still have problems in the following aspects: First, traditional recommendation systems are difficult to adapt to the dynamic understanding of complex intentions in conversational interactions, resulting in the lack of pertinence and flexibility of recommendation results; Second, the combination method of existing LLMs and recommendation systems is still in the exploration stage, and how to balance computational cost and real-time performance while improving the user experience remains a major challenge. Therefore, developing a new generation of recommendation systems that can fully combine the advantages of LLMs and effectively address the needs of dynamic interactions has become the key research direction.

[0006] In the field of vocational education, personalized learning assistants (PLAs), as intelligent tools, aim to provide customized learning paths, resource recommendations, and real-time feedback based on learners' backgrounds, learning goals, and progress. Existing personalized learning assistants usually rely on recommendation systems, intelligent tutoring systems, and knowledge graph technologies, and recommend suitable learning content for learners by analyzing their historical behaviors and preferences. These technologies have improved learning efficiency and learning experience to a certain extent, especially showing important value in massive open online courses (MOOCs) and vocational skills training. However, existing technologies still have obvious defects: 1. Insufficient understanding of learning intentions: Traditional personalized learning assistants mainly rely on fixed rules or recommendations based on historical behaviors, lacking a deep understanding of learners' dynamic learning needs. For example, when learners' interests or goals change, the system is difficult to quickly adjust the recommendation strategy, resulting in an inflexible learning path; 2. Limited dialogue interaction ability: Existing personalized learning assistants usually present one-way content recommendations or preset Q&A forms in interactions with learners, lacking the multi-round dialogue ability similar to human tutors and unable to effectively respond to complex questions or dynamic learning needs raised by learners; 3. Insufficient accuracy of personalized recommendation: Although some systems combine knowledge graphs and deep learning technologies, their comprehensive modeling capabilities for learners' personality characteristics (such as cognitive level, learning habits) and multimodal data (such as speech, text, video) are still limited, making it difficult to further improve the accuracy of personalized recommendations. 4. Weak cross - domain transfer ability: In the vocational education scenario, learners may need to improve skills across domains, for example, switching from mechanical operation to intelligent manufacturing. Existing technologies lack in - depth modeling of the relationships between knowledge in different domains and are difficult to provide effective cross - domain learning suggestions for learners. 5. Lack of real - time performance and adaptability: Vocational education often requires learners to master specific skills within a short period. However, existing personalized learning assistants usually cannot capture the current state of learners in real - time and dynamically adjust the learning path, resulting in a decline in learning efficiency.

[0007] In summary, existing personalized learning assistants have technical problems such as insufficient personalization and lack of interactivity. Especially in vocational education, they face the challenge of being unable to provide dynamically adjusted and customized learning content according to the actual needs of students. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to provide a method for constructing a personalized learning assistant for vocational education based on virtual digital humans, which can realize dynamic and personalized learning path recommendation and real - time feedback.

[0009] The technical solution adopted by the present invention to solve the above - mentioned technical problem is: A method for constructing a personalized learning assistant for vocational education based on virtual digital humans, comprising the following steps: Step 1. Set up a course library list Pre - set a course library list for recording relevant information of each course in vocational education for users. The relevant information includes the degree of user preference X for the course, the learning progress P of the user for the course, and the degree of user mastery S of the course. Each index is arranged from small to large according to the degree level, and a degree value corresponding to each index is obtained by assigning a number between 0 and 100. Step 2. Update the course library list Input the user information related to vocational education into the large language model for processing, and output the course name, the degree values of its various indicators, and the user intention. If the output course name does not exist in the course library list preset in step 1, add the course name and the degree values of its various indicators to the course library list. If the output course name exists in the course library list preset in step 1, re-calculate the degree values of each indicator of the course for the user in the course library list. The calculation method is as follows: G = tGold+(1 - t)Gnew, where G refers to the degree value of each indicator of the course for the user in the updated course library list, Gold refers to the degree value of each indicator of the course for the user in the preset course library list, Gnew refers to the degree value of each indicator of the course for the user output in step 2, and t is a weight coefficient used to control the weight of the old and new values, with a value of 0 < t < 1; Step 3: Retrieval of learning resources Form the text by combining the course name, the degree values of its various indicators, and the user intention output in step 2, and input it into the Text2SQL model. The Text2SQL model generates a query SQL, and uses the query SQL to retrieve in the database related to vocational education to obtain a recommended list of learning resources, including resource name, resource link, and the course to which the resource belongs; Step 4: Re-ranking of the learning resource list Perform a correlation analysis on the learning resource list obtained in step 3 and the updated course library list obtained in step 2 using a decision tree model, filter the learning resource list with a correlation coefficient greater than 0.7, input the filtered learning resource list and the user information related to vocational education in step 2 into a cross-encoder for processing, obtain a recommended list of re-ranked learning resources, and complete the construction of a personalized learning assistant for vocational education.

[0010] Furthermore, the weight coefficient t described in step (2) is a dynamically adjusted coefficient, which is dynamically adjusted according to the user's recent behavior frequency and activity. The formula is as follows: t = α ⋅ Behavior frequency+(1 - α )⋅ Default weight, where a = 0.6, the behavior frequency is the percentage of the number of times the user asks questions about this course within a week in the total number of times all users ask questions about this course, and the default weight is 0.5. Introducing a dynamically adjusted weight coefficient can update the course library list in a timely manner according to the user's behavior changes, avoid problems such as update lag or excessive smoothing caused by a fixed weight, be closer to the user's real-time needs, and thus improve the personalization and accuracy of recommendations.

[0011] Furthermore, the learning resource list described in step (3) includes resource name, resource link, and the course to which the resource belongs.

[0012] Compared with the prior art, the advantages of the present invention are: 1. Personalized learning experience: Dynamically adjust recommended content based on user needs and feedback to ensure a high degree of match between learning resources and personal goals and interests, enhancing learning effectiveness. 2. Improve learning efficiency: Through precise recommendations, users can quickly find the most relevant and useful resources, reducing time waste and enhancing learning efficiency. 3. Enhance user engagement: Personalized recommendations can stimulate users' learning interests and increase their sense of participation and enthusiasm during the learning process. 4. Dynamic adaptability: The system can continuously optimize recommendations based on users' real-time feedback and learning progress to ensure that resources always meet users' current needs and levels. 5. Improve learning outcomes: By providing targeted learning resources, help users better master knowledge and skills, thereby improving the overall effectiveness of vocational education. 6. Enhance user satisfaction: By providing learning resources that meet users' expectations, improve users' satisfaction and loyalty to the vocational education platform.

[0013] In summary, the method for constructing a personalized learning assistant for vocational education based on virtual digital humans in the present invention adopts large language model technology. Through the accurate identification of students' learning progress, knowledge mastery, and personalized needs, it realizes intelligent adjustment of learning content and real-time feedback. The combination of these technologies can ensure that the virtual digital human can not only provide rich knowledge imparting but also interact flexibly according to students' feedback, thus effectively enhancing the pertinence and effectiveness of learning, effectively solving the problems of insufficient personalized learning and interactivity in traditional vocational education, enhancing learners' sense of participation and immersion, and being able to provide accurate, detailed, and context-rich customized responses according to students' needs, thereby improving learning effects and learning experiences. Detailed implementation manners

[0014] The present invention will be further described in detail below in conjunction with embodiments.

[0015] Specific embodiment: A method for constructing a personalized learning assistant for vocational education based on the LLM model includes the following steps: Step 1. Set up a course library list Pre-set a course library list for recording relevant information of each course in vocational education for users. The relevant information includes the degree of user preference X for the course, the learning progress P of the user for the course, and the mastery degree S of the user for the course. Each index is arranged from small to large according to the degree level, and a degree value corresponding to each index is obtained by assigning a number between 0 and 100.

[0016] Step 2. Update the course library list Input the user information related to vocational education into a large language model for processing, and output the course name, the degree values of its various indicators, and the user's intentions (seeking explanations, finding resources, modifying and paraphrasing text (polish and modify the article I wrote), seeking questions (give me some practice questions), and other unclassified questions, etc.). The large language model used above is not limited to a single type of LLM, and both open-source models and commercial models can be used (such as Tongyi Qianwen); If the course name output does not exist in the course library list preset in step 1, add the course name and the degree values of its various indicators to the course library list. If the course name output exists in the course library list preset in step 1, re-calculate the degree values of each indicator of the course for the user in the course library list. The calculation method is as follows: G = tGold+(1 - t)Gnew, where G refers to the degree value of each indicator of the course for the user in the updated course library list, Gold refers to the degree value of each indicator of the course for the user in the preset course library list, Gnew refers to the degree value of each indicator of the course for the user output in step 2, and t is a weight coefficient used to control the weight of the old and new values, with a value of 0 < t < 1; Among them, the weight coefficient t is a dynamically adjusted coefficient, which is dynamically adjusted according to the user's recent behavior frequency and activity. The formula is as follows: t = α ⋅ Behavior frequency+(1 - α )⋅ Default weight, where a = 0.6, the behavior frequency is the percentage ranking of the number of questions asked by the user about this course within a week among the number of questions asked by all users about this course, and the default weight is 0.5. Introducing a dynamically adjusted weight coefficient can update the course library list in a timely manner according to the user's behavior changes, avoid problems such as update lag or excessive smoothing caused by fixed weights, be closer to the user's real-time needs, and thus improve the personalization and accuracy of recommendations.

[0017] Step 3: Retrieval of learning resources After forming the text with the course name, the degree values of its various indicators, and the user's intentions output in step 2, input it into the Text2SQL model. The Text2SQL model generates a query SQL, and use the query SQL to retrieve in the vocational education-related database to obtain a recommended list of learning resources, including resource names, resource links, and the courses to which the resources belong.

[0018] Step 4: Re-ranking of the learning resource list Perform a correlation analysis on the learning resource list obtained in step 3 and the updated course library list obtained in step 3 using a decision tree model, filter the learning resource list with a correlation coefficient greater than 0.7, and input the filtered learning resource list and the vocational education-related user information in step 2 into a cross-encoder for processing to obtain a recommended re-ranked learning resource list, thus completing the construction of a personalized learning assistant for vocational education.

[0019] The above description is not a limitation of the present invention, nor is the present invention limited to the above examples. Changes, modifications, additions or substitutions made by those of ordinary skill in the art within the scope of the essence of the present invention shall also fall within the protection scope of the present invention.

Claims

1. A method for constructing a personalized learning assistant for vocational education based on virtual digital humans, characterized in that The following steps are involved: Step 1: Set up the course library list A course library list is pre-set to record the user's information about each course in vocational education. The relevant information includes the user's preference for the course (X), the user's learning progress (P), and the user's mastery of the course (S). Each indicator is arranged in ascending order of degree, and a number between 0 and 100 is assigned to obtain the corresponding degree value of each indicator. Step 2: Update the course library list Input the user information related to vocational education into the large language model for processing, and output the course name and the degree value of each indicator and the user intention. If the output course name does not exist in the course library list preset in step 1, then the course name and the degree value of each indicator are added to the course library list. If the output course name exists in the course library list preset in step 1, then the degree value of each indicator of the course in the course library list is re-assigned and calculated. The calculation method is as follows: G=tGold+(1-t)Gnew, G refers to the degree value of each indicator of the course in the updated course library list, Gold refers to the degree value of each indicator of the course in the preset course library list, Gnew refers to the degree value of each indicator of the course output by the user in step 2, and t value is 0 <t<1; Step 3: Retrieve learning resources The course name, the degree values of its various indicators, and the user intent output in step 2 are combined into a text and then input into the Text2SQL model. The Text2SQL model generates a query SQL statement, which is used to search the vocational education-related database to obtain a list of recommended learning resources, including the resource name, resource link, and the course to which the resource belongs. Step 4: Reorder the learning resource list The learning resource list obtained in step 3 and the updated course library list obtained in step 2 are subjected to correlation analysis using a decision tree model, and learning resource lists with correlation coefficients greater than 0.7 are screened. The screened learning resource list and the vocational education-related user information in step 2 are input into the cross-encoder for processing to obtain the re-sorted learning resource list recommendation, thus completing the construction of a personalized learning assistant for vocational education.

2. The method for constructing a personalized learning assistant for vocational education based on a virtual digital human according to claim 1, wherein: The weight coefficient t described in step (2) is a dynamic adjustment coefficient, which is dynamically adjusted according to the user's recent behavior frequency and activity. The formula is as follows: t = α ⋅ Behavior frequency + (1 - α ) ⋅ Default weight, where a = 0.

6. The behavior frequency is the percentage ranking of the number of questions asked by the user about the course within one week among the number of questions asked by all users about the course. The default weight is 0.

5.

3. A method for constructing a personalized learning assistant for vocational education based on a virtual digital human according to claim 1, characterized in that: The learning resource list described in step (3) includes the resource name, resource link and the course to which the resource belongs.