Intelligent accompanying learning method and device, storage medium and equipment

Abstract

The invention discloses an intelligent accompanying learning method and device, a storage medium and equipment, and the method comprises the steps: firstly obtaining first learning information and second learning information of a target user, and then employing the first learning information of the target user, combining with a first prompt, inputting the first learning information into an LLM, and obtaining a user portrait of the target user outputted by a model; inputting the user portrait of the target user, the second learning information of the target user and the agent information into the LLM in combination with the second prompt to obtain a learning strategy for the target user output by the model; and according to the learning strategy, collaborative calling is carried out on the intelligent agent through LLM, and intelligent accompanying learning of the target user is realized. Therefore, the timeliness and accuracy of the whole accompanying learning process are effectively improved, the accompanying learning effect of the target user is effectively improved, and the learning experience of the target user is also improved.

Description

Technical Field

[0001] The present application relates to the field of natural language processing technology, and in particular to an intelligent learning companion method, device, storage medium and equipment. Background Art

[0002] With the rapid development of educational technology, people are increasingly demanding personalized learning and teaching students in accordance with their aptitude. However, many existing learning assistance methods have not really met users' learning demands in a timely manner and have not achieved the goal of reducing learning burden and improving learning efficiency.

[0003] At present, there are usually three methods of assisting users in learning: the first is to use self-search resources / tool-type learning products for assisted learning; the second is to use personalized assisted learning tool products with learning situation modeling for assisted learning; the third is the synchronous planning learning assisted learning method based on expert knowledge. However, none of these existing methods of assisting users in learning truly have the concept of learning companionship. They only reflect the companionship of a cold training machine, rather than playing the role of a learning partner who tailors learning plans for users and answers questions in a timely manner during the process. Therefore, the ideal learning companionship effect cannot be achieved, and the user's learning experience will also be reduced. Summary of the invention

[0004] The main purpose of the embodiments of the present application is to provide an intelligent learning companion method, device, storage medium and equipment, which can first build a personalized portrait of the user based on a large language model (LLM), and then use the process self-reflection mechanism to effectively improve the learning companion effect for the user, thereby enhancing the user's learning experience.

[0005] The present application embodiment provides an intelligent learning companion method, including: Acquire first learning information and second learning information of a target user; Using the first learning information of the target user, combined with the first prompt instruction prompt, inputting it into the large language model, and obtaining a user portrait of the target user output by the model; Utilizing the user portrait of the target user, the second learning information of the target user, and the agent information, combined with the second prompt instruction prompt, inputting them into the large language model, and obtaining a learning strategy for the target user output by the model; According to the learning strategy, the agent is collaboratively called through the large language model to achieve companion learning for the target user.

[0006] In a possible implementation, the first learning information of the target user includes the current learning state of the target user, the historical learning information of the target user, and the scene information of the target user; the first learning information of the target user is used in combination with the first prompt instruction prompt, and is input into the large language model to obtain the user portrait of the target user output by the model, including: Structuring the historical learning information of the target user, and using the obtained processing result in combination with the first prompt instruction prompt, inputting it into the large language model to obtain a historical portrait of the target user output by the model; Using the current learning state of the target user, the historical portrait of the target user, and the scene information of the target user, combined with the updated first prompt instruction prompt, input them into the large language model to obtain the instant portrait of the target user output by the model; The overall user profile of the target user is constructed by using the historical profile of the target user and the instant profile of the target user.

[0007] In a possible implementation, the second learning information of the target user includes the learning needs proposed by the target user; the learning needs proposed by the target user are determined based on at least one round of conversation between the large language model and the target user.

[0008] In a possible implementation, the user profile of the target user, the second learning information of the target user, and the agent information are combined with the second prompt instruction prompt, input into the large language model, and the learning strategy for the target user output by the model is obtained, including: The user portrait of the target user, the learning needs proposed by the target user, the scene information of the target user and the agent information are used, combined with the second prompt instruction prompt, and input into the large language model to obtain the learning strategy for the target user output by the model.

[0009] In a possible implementation, the step of collaboratively calling the agent through the large language model according to the learning strategy to achieve companion learning for the target user includes: According to the learning strategy, each intelligent agent is called through the large language model to realize the main process of companion learning for the target user; and in the process of realizing the detailed process of companion learning, the mutual calling between each intelligent agent is realized to realize the coordinated calling of each intelligent agent, so as to complete the whole companion learning processing process for the target user.

[0010] In a possible implementation, the method further includes: In the learning process, based on the feedback information of the target user, the user profile of the target user is updated by using the process self-reflection mechanism; The updated user portrait of the target user is used to update the learning strategy of the target user, and the updated learning strategy is used to collaboratively call the intelligent agent again through the large language model to achieve intelligent learning companionship for the target user until the learning needs proposed by the target user are met.

[0011] In a possible implementation, in the companion learning process, according to the feedback information of the target user, a process self-reflection mechanism is used to update the user profile of the target user, including: In the learning process, the target learning resources to be recommended are determined according to the learning strategy; and it is determined whether the relevance between the target learning resources and the learning needs and user profile of the target user meets the preset requirements; If so, the target learning resource is recommended to the target user, and real-time interaction is performed with the target user based on the target user's learning participation, interaction during participation, and various learning behaviors and learning attitudes, so as to recommend the adjusted target learning resource to the target user and update the relevant content in the user profile of the target user based on the target user's feedback information; If not, confirm the adjusted target learning resource through secondary interaction with the target user; recommend the adjusted target learning resource to the target user, and update the relevant content in the user profile of the target user; The learning results of the target user on the target learning resource and / or the adjusted target learning resource are analyzed, and the user portrait of the target user is updated according to the analysis results.

[0012] In a possible implementation, the analysis result includes an analysis result of the target user's insufficient completion of the target learning resource and / or the learning content of the target learning resource and the cause of the wrong questions.

[0013] In one possible implementation, the current learning status of the target user includes at least one of the current expression of the target user, the text proposed and / or the language and behavioral expressions issued by the target user; the historical learning information of the target user includes at least one of the historical behavior sequence, interaction sequence, answer content, process performance and results of the target user; the scene information of the target user includes at least one of the review scene, preview scene, exam preparation scene, and holiday learning scene.

[0014] The present application also provides an intelligent learning companion device, including: An acquisition unit, used for acquiring first learning information and second learning information of a target user; A first input unit is used to use the first learning information of the target user, combined with a first prompt instruction prompt, to input into the large language model to obtain a user portrait of the target user output by the model; A second input unit is used to use the user portrait of the target user, the second learning information of the target user and the agent information, combined with the second prompt instruction prompt, to input into the large language model to obtain the learning strategy for the target user output by the model; A companion learning unit is used to collaboratively call the intelligent agent through the large language model according to the learning strategy to achieve companion learning for the target user.

[0015] In a possible implementation, the first learning information of the target user includes the current learning state of the target user, the historical learning information of the target user, and the scene information of the target user; the first input unit includes: The first input subunit is used to perform structured processing on the historical learning information of the target user, and use the obtained processing result in combination with the first prompt instruction prompt to input it into the large language model to obtain the historical portrait of the target user output by the model; The second input subunit is used to utilize the current learning state of the target user, the historical portrait of the target user, and the scene information of the target user, combined with the updated first prompt instruction prompt, and input them into the large language model to obtain an instant portrait of the target user output by the model; The construction subunit is used to construct an overall user profile of the target user by using the historical profile of the target user and the instant profile of the target user.

[0016] In a possible implementation, the second learning information of the target user includes the learning needs proposed by the target user; the learning needs proposed by the target user are determined based on at least one round of conversation between the large language model and the target user.

[0017] In a possible implementation, the second input unit is specifically used to: The user portrait of the target user, the learning needs proposed by the target user, the scene information of the target user and the agent information are used, combined with the second prompt instruction prompt, and input into the large language model to obtain the learning strategy for the target user output by the model.

[0018] In a possible implementation, the companion learning unit is specifically used for: According to the learning strategy, each intelligent agent is called through the large language model to realize the main process of companion learning for the target user; and in the process of realizing the detailed process of companion learning, the mutual calling between each intelligent agent is realized to realize the coordinated calling of each intelligent agent, so as to complete the whole companion learning processing process for the target user.

[0019] In a possible implementation manner, the device further includes: An updating unit, used to update the user profile of the target user by using a process self-reflection mechanism according to the feedback information of the target user during the companion learning process; The calling unit is used to use the updated user portrait of the target user to update the learning strategy of the target user, and use the updated learning strategy to collaboratively call the intelligent agent again through the large language model to achieve intelligent learning companionship for the target user until the learning needs proposed by the target user are met.

[0020] In a possible implementation, the updating unit includes: A judgment subunit is used to determine the target learning resources to be recommended according to the learning strategy during the companion learning process; and to judge whether the relevance between the target learning resources and the learning needs and user profile of the target user meets the preset requirements; A first updating subunit is used for recommending the target learning resource to the target user if it is determined that the relevance between the target learning resource and the learning needs and user profile of the target user meets preset requirements, and for interacting with the target user in real time according to the learning participation, interaction during the participation process, and various learning behaviors and learning attitudes of the target user, so as to recommend the adjusted target learning resource to the target user according to the feedback information of the target user, and to update the relevant content in the user profile of the target user; The second updating subunit is used to confirm the adjusted target learning resource through secondary interaction with the target user if it is determined that the relevance between the target learning resource and the learning needs and user profile of the target user does not meet the preset requirements; and recommend the adjusted target learning resource to the target user, and update the relevant content in the user profile of the target user; The third updating subunit is used to analyze the learning results of the target user on the target learning resource and / or the adjusted target learning resource, and update the user portrait of the target user according to the analysis results.

[0021] In a possible implementation, the analysis result includes an analysis result of the target user's insufficient completion of the target learning resource and / or the learning content of the target learning resource and the cause of the wrong questions.

[0022] In one possible implementation, the current learning status of the target user includes at least one of the current expression of the target user, the text proposed and / or the language and behavioral expressions issued by the target user; the historical learning information of the target user includes at least one of the historical behavior sequence, interaction sequence, answer content, process performance and results of the target user; the scene information of the target user includes at least one of the review scene, preview scene, exam preparation scene, and holiday learning scene.

[0023] The embodiment of the present application also provides an intelligent learning companion device, including: a processor, a memory, and a system bus; The processor and the memory are connected via the system bus; The memory is used to store one or more programs, and the one or more programs include instructions. When the instructions are executed by the processor, the processor executes any one of the implementation methods of the above-mentioned intelligent companion learning method.

[0024] An embodiment of the present application also provides a computer-readable storage medium, in which instructions are stored. When the instructions are executed on a terminal device, the terminal device executes any one of the implementation methods of the above-mentioned intelligent companion learning method.

[0025] The embodiment of the present application also provides a computer program product. When the computer program product is run on a terminal device, the terminal device executes any one of the implementation methods of the above-mentioned intelligent companion learning method.

[0026] An intelligent learning companion method, apparatus, storage medium and device provided in the embodiments of the present application first obtain the first learning information and the second learning information of the target user; then use the first learning information of the target user, combined with the first prompt instruction prompt, input it into the LLM, and obtain the user portrait of the target user output by the model; then use the user portrait of the target user, the second learning information of the target user and the intelligent agent information, combined with the second prompt instruction prompt, input it into the LLM, and obtain the learning strategy for the target user output by the model; then according to the learning strategy, the intelligent agent is collaboratively called through the LLM to realize learning companion for the target user.

[0027] It can be seen that since the present application first utilizes the target user's first learning information (such as the target user's historical learning information, current learning status, and scene information) based on LLM, a full-scene portrait of the target user is accurately constructed, and then the learning strategy is determined based on the full-scene portrait of the target user and the target user's second learning information (such as the learning needs proposed by the target user) and the intelligent agent information. Based on the learning strategy, through the collaborative calling of the intelligent agent combined with the process self-reflection mechanism, intelligent learning companionship for the target user is realized, which effectively improves the timeliness and accuracy of the entire learning companionship process, thereby improving the learning companionship effect for the target user, and then improving the learning experience of the target user. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0029] Figure 1 A flowchart of an intelligent learning companion method provided in an embodiment of the present application; Figure 2 A schematic diagram of the overall implementation process of the intelligent learning companion provided in the embodiment of the present application; Figure 3 An example diagram of a process for updating a user portrait of a target user by using a process self-reflection mechanism based on feedback information from the target user provided in an embodiment of the present application; Figure 4 A schematic diagram of the composition of an intelligent learning companion device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0030] With the rapid development of educational technology, more and more auxiliary learning products are appearing in people's lives to meet people's demands for personalized learning and teaching students in accordance with their aptitude. However, many current products that assist users in learning still tend to allocate fixed learning resources to users under a given learning outline. Of course, with the evolution of technology, personalized recommendations for learning resources have been achieved through modeling historical learning situations, but basically, current auxiliary learning products are passively accepted by users. There is no free interaction and timely response or insight into the real needs of users, and it is impossible to understand the emotions of users in the learning process, which makes users feel that the learning process is very boring.

[0031] Specifically, products that assist users in learning are basically concentrated in the following three categories: Category 1: Search tools / resources. This type of learning product mainly provides more resources to facilitate users to screen and train themselves. The product features rich resources and requires users to have a strong sense of independent screening. The corresponding technical solution is mainly based on tag search.

[0032] Category 2: Auxiliary learning tools. This type of method is mainly based on the user's historical learning sequence, modeling the user's learning profile, and then predicting the user's mastery of future knowledge points through probability, so as to recommend subsequent learning paths. The focus is on prediction technology and students' frequent use to achieve rich historical learning.

[0033] Category 3: Synchronous planning. The main idea is to distribute relevant resources to users regularly according to the teaching syllabus or relevant teaching progress. The process may be adaptively adjusted based on user levels, but the essence is still a combination of resource allocation and tool scheduling. The overall perception given to users is to complete tasks according to plan.

[0034] It can be seen that the current existing products that assist users in learning are basically concentrated in the categories of self-search resources / tools, personalized learning modeling, and planning based on expert knowledge, and the concept of learning companionship has not yet really emerged. The learning companionship here is more about learning companionship, which enhances the user's learning confidence and sense of accomplishment during the learning process, and makes users feel that learning products (such as learning machines) are not cold training machines but learning partners who understand me, tailor learning plans for me, and answer questions and resolve doubts in a timely manner during the process. Therefore, the existing methods and products that assist users in learning cannot achieve the ideal learning companionship effect, and will also reduce the user's learning experience.

[0035] In addition, the applicant has also found that in recent years, large language models (LLMs) have provided powerful reasoning and outstanding performance in processing complex human emotions and unstructured interactions, breaking the bottleneck of traditional deep learning / machine learning that cannot reach human-likeness. At the same time, with the rapid development of intelligent agent technology under various LLMs, it is not only possible to dynamically adjust teaching strategies according to students' learning conditions and feedback, but also to provide more humane learning support by simulating the behavior of human teachers. Intelligent agents can also better integrate and analyze various data in the students' learning process, such as learning habits, preferences, and performance, thereby providing teachers (such as LLMs) with deeper teaching insights. This data-driven approach can help teachers (such as LLMs) formulate teaching plans more effectively and provide personalized teaching based on students' needs.

[0036] Therefore, the present application proposes to combine advanced LLM technology to achieve a more humane and intelligent learning process for users to solve the above defects. Specifically, the present application provides an intelligent learning companion method, which first obtains the first learning information and second learning information of the target user; then uses the first learning information of the target user, combined with the first prompt instruction prompt, inputs it into the LLM, and obtains the user portrait of the target user output by the model; then uses the user portrait of the target user, the second learning information of the target user, and the intelligent agent information, combined with the second prompt instruction prompt, inputs it into the LLM, and obtains the learning strategy for the target user output by the model; then, according to the learning strategy, the intelligent agent is collaboratively called through the LLM to achieve learning companionship for the target user.

[0037] It can be seen that since this application first uses the target user's first learning information (such as the target user's historical learning information and current learning status, as well as the scene information, etc.) based on LLM, it accurately constructs a full-scene portrait of the target user, and then determines the learning strategy based on the target user's full-scene portrait and the target user's second learning information (such as the learning needs proposed by the target user, etc.) and the intelligent agent information, and based on the learning strategy, through the collaborative call of the intelligent agent combined with the process self-reflection mechanism, intelligent learning companionship for the target user is realized, which effectively improves the timeliness and accuracy of the entire learning companionship process, thereby improving the learning companionship effect for the target user, and then improving the target user's learning experience.

[0038] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0039] First embodiment

[0040] See also Figure 1 , is a flow chart of an intelligent learning companion method provided in this embodiment, the method comprising the following steps: S101: Acquire first learning information and second learning information of a target user.

[0041] In this embodiment, any user (such as a student, a doctor, etc.) who uses the method provided by this embodiment to realize intelligent learning companionship is defined as a target user, and in order to realize a more humane and intelligent learning companionship process for the target user to improve the learning experience of the target user, this application first interacts with the target user in real time in a single round or multiple rounds based on a large language model (LLM), so as to obtain the learning needs proposed by the target user (i.e., disassemble the specific intention of the target user) through the text and / or language request (query), behavior process, expression feedback, etc. proposed by the target user during the interaction process, so as to form the second learning information of the target user. And obtain the current learning status of the target user, the historical learning information of the target user, and the scene information of the target user, so as to form the first learning information of the target user, so as to execute the subsequent steps S102-S105.

[0042] It should be noted that this application does not limit the content and acquisition method of the learning needs proposed by the target user, which can be selected or set according to actual conditions and experience. An optional implementation method is to obtain the learning needs proposed by the target user based on at least one round of dialogue between the large language model (LLM) and the target user.

[0043] In this implementation, the learning needs proposed by the target users (represented as ) One possible source is to directly obtain simple learning needs based on a single round of query between LLM and the target user; another complex learning need may be extracted from multiple rounds of dialogue between LLM and the target user, which requires condensing the intentions of the historical multiple rounds of dialogue between LLM and the target user. Another more complex learning need may be analyzed in combination with multi-modal behavior states. Here, this application combines the target user's learning behavior, multiple rounds of interaction, answering status and other multi-dimensional information in the learning machine scenario as an example to accurately determine the target user's learning needs. The calculation formula is as follows:

[0044] Among them, Router() indicates that it is necessary to combine the actual effect and the scene information of the target user, and self-consistently select different methods through LLM (or other traditional small models such as pre-trained language representation models) To improve the generation diversity and effect. Multi_feature represents multi-dimensional features, and the specific content is not limited. For example, Multi_feature can be represented as the following example content: {“User needs”: “I need a set of pre-exam papers suitable for me”, “Current Profile”: “Highly afraid of difficulties, weak foundation, tired of video learning...”, "Learning behavior": "Prefer objective questions, average concentration time is less than 30 minutes...", "History Interaction": "", "Answer Status": "Simple question accuracy rate xx, difficult question accuracy rate xx, different question types accuracy rate xx...", }.

[0045] In addition, it should be noted that the present application does not limit the content and acquisition method of the current learning status of the target user, which can be selected or set according to the actual situation and experience. An optional implementation method is that the current learning status of the target user may include but is not limited to at least one of the current expression of the target user, the text proposed by the target user (such as the target user inputting the text "too difficult") and / or the language and behavior expression issued.

[0046] Furthermore, the present application does not limit the content and acquisition method of the target user's historical learning information, which can be selected or set according to actual conditions and experience. An optional implementation method is that the target user's historical learning information may include but is not limited to at least one of the target user's historical behavior sequence, interaction sequence, answer content, process performance, and results.

[0047] Furthermore, this application does not limit the content and acquisition method of the scene information of the target user, which can be selected or set according to the actual situation and experience. An optional implementation method is that the scene information of the target user can include but is not limited to at least one of the review scene, preview scene, test preparation scene, and holiday (such as winter and summer vacation) learning scene, where each scene has a different learning focus, such as the review scene mainly focuses on students' weaknesses and easy mistakes, and the key and difficult points of the exam.

[0048] S102: Utilize the first learning information of the target user, combined with the first prompt instruction prompt, and input it into the large language model to obtain a user portrait of the target user output by the model.

[0049] In this embodiment, it should be noted that in order to achieve a more humane and intelligent learning process for the target user to improve the learning experience of the target user, after obtaining the first learning information of the target user (including but not limited to the current learning status of the target user (such as the current expression of the target user, the text proposed and / or the language and behavior expression issued, etc.), the historical learning information of the target user (such as historical behavior sequence, interaction sequence, answer content, process performance and results, etc.) and the scene where the target user is located (such as review scene, preview scene, test preparation scene, holiday learning scene, etc.) information) through step S101, it can be further based on the LLM technology to convert the information that is difficult for other traditional small models such as pre-trained language representation models to learn. The processed unstructured information is structured, and a personalized user portrait of the target user is constructed based on the full-scenario multimodal learning features contained in the processing results. Specifically, the current learning status of the target user, the historical learning information and the scene information of the target user included in the first learning information of the target user can be integrated into the prompt instruction prompt (here defined as the first prompt instruction), and input into the large language model (LLM) to obtain a personalized portrait of the target user output by the large language model. In addition, in order to enrich the portrait content of the target user, the traditional portrait features of the target user (such as basic attribute features, statistical attribute features, etc.) can also be combined to form an overall user portrait of the target user for executing the subsequent step S103.

[0050] Among them, this application does not limit the content and classification of the target user's portrait. For example, the target user's portrait content can be divided into basic attributes, statistical attributes and intrinsic attributes. Among them, basic attributes refer to the target user's personal information such as age, gender, geographic location, occupation, etc.; statistical attributes can be calculated from teaching data based on set statistical indicators. For example, the average accuracy of target users' questions can be counted from static data, or the average viewing time of target users' online courses can be counted based on dynamic data; intrinsic attributes refer to hidden attributes that need to be mined based on various learning data using expert knowledge, deep learning and other methods, such as students' ability status, emotional status, etc. This requires special modeling and selection of data. For example, this application can use the following paradigm when measuring the target user's mastery of a certain knowledge point and learning emotion: Mastery = KT (historical answer sequence, test characteristics, score) Learning emotion = LM (historical interaction data, learning behavior characteristics) Among them, LM represents language model, then "learning sentiment = LM (historical interaction data, learning behavior characteristics)" means that the historical interaction data and learning behavior characteristics of the target user are jointly analyzed through the language model LM to generate or predict the target user's learning sentiment; KT represents knowledge tracing model, then "mastery = KT (historical answer sequence, test question characteristics, score)" means that through the knowledge tracing model KT, combined with the target user's historical answer sequence (such as historical answer mode), test question characteristics (such as question attributes) and scores (such as score details), the real-time mastery of specific knowledge points is dynamically calculated.

[0051] In addition, in order to make up for the deficiency of simply using cognitive abilities such as intelligence and grades to reflect the learning success of target users (such as students), this application proposes to include non-cognitive abilities such as perseverance and self-control, and non-intellectual factors such as learning motivation, self-awareness, and sound personality as part of the important user portrait content; such features cannot be effectively modeled by traditional small models such as pre-trained language representation models, such as the inability to understand user behavior, answer results, process emotions, interaction data, etc.

[0052] Therefore, an optional implementation method is, after obtaining unstructured information such as the historical learning information of the target user within a period of time in the past (such as historical behavior sequence, interaction sequence, answer content, process performance and results, etc.), the historical learning information of the target user within this period of time can be further structured, and the processing result can be integrated into the first prompt instruction prompt, input into the large language model, and obtain the historical portrait of the target user output by the model.

[0053] In this implementation, the structural processing of the target user's historical learning information refers to structural processing according to different specific tasks. The specific implementation method is not restricted. For example, the historical learning information data of students (as an example of the target user) can be uniformly converted into json format. The example of converting each category of data into a list and a string through the model can be as follows: { "Answer sequence": [(Question 1, 0.8), (Question 2, 1.0), ...], “Process performance”: [(Test 1, normal), (Test 2, frowning), ...], }.

[0054] On this basis, the above processing results are integrated into the prompt instructions, and an example of inputting the large language model can be: "As a subject teacher with rich teaching experience, please be able to conduct step-by-step analysis on students' {"historical structured characteristics"} from the perspective of students' learning content, problem-solving performance, and emotional expression during the process..., etc., to form relevant portrait features. The portrait features include "students' initiative, sense of purpose, current emotions, usual thinking depth..."

[0055] It can be understood that this implementation method structures the historical learning information content of the target user, such as the historical behavior sequence, interaction sequence, answer content, process performance and results, and then uses the natural language understanding ability and thinking logic reasoning ability of LLM itself to perform multi-link chain of thought (Chain of Thought, COT) reasoning analysis. For example, learning motivation is analyzed through LLM to analyze the students' (as an example of the target user) participation frequency, completion status and the seriousness of the content completion process of the answer process (such as the richness of the notes and answers after optical character recognition (OCR), process attitude, clarity of answer ideas, depth of thinking, etc.) and various learning behaviors after the answer results. After giving the expert constraint framework, let LLM conduct traceability reasoning analysis based on various dimensions and data to determine whether the students (as an example of the target user) have strong initiative and sense of purpose.

[0056] In addition, another optional implementation method is, after obtaining the historical portrait of the target user (such as the historical portrait at the previous moment), the current learning status of the target user (such as the current expression of the target user, the text proposed and / or the language and behavioral expressions issued, etc.), and the scene information of the target user (such as review scene, preview scene, exam preparation scene, holiday learning scene, etc.), the current learning status of the target user, the historical portrait at the previous moment and the scene information in which the target user is located can be further integrated into the updated first prompt instruction prompt, input into the large language model, and obtain the instant portrait of the target user output by the model.

[0057] In this implementation, the target user (such as a student) has immediate needs and emotional states in the learning process, such as fatigue and poor emotional state. These immediate portrait captures are also crucial for accurately determining the learning companion strategy for the target user in the subsequent steps. Therefore, this application is based on the joint application of LLM and related plug-ins, and captures the target user (such as a student) at each moment (represented here as ), the historical portrait of the previous moment (represented here as ) and the scene information of the target user (represented here as ), integrated into the prompt instruction (represented as P here) as the input of LLM, so as to form a real-time generation of the target user's portrait through the natural language understanding ability and thinking logic reasoning ability of LLM itself, and obtain the target user's instant portrait (represented as ), the specific calculation formula is as follows:

[0058] Among them, the purpose of the prompt instruction (P) is to generate a user portrait at a certain moment. Its main function is to integrate relevant scenarios and portrait features into relevant prompt words with thought chains according to hyperparameters. An example can be: "As a teacher with rich teaching experience, please combine the current learning {status}, students' {historical portrait features} and the current students' {scenario information}, and gradually combine and analyze the students' latest portrait features, generate features for updating existing portrait dimensions, and if new requirements are involved, the description needs to be condensed into [current requirements]."

[0059] and It refers to a structured portrait, which updates some feature values ​​of the existing historical portrait. The following example only shows the changed features. The changed feature values ​​will affect the recommendation and dialogue strategy of the subsequent steps. For example, if a student has multiple difficult questions at a certain moment, then An example of could be: { … "Fear of difficulty": low-->high, "Abnormal behavior": frequently skipping questions, … }.

[0060] Furthermore, after obtaining the historical portrait and the instant portrait of the target user, the two can be combined with the traditional portrait features of the target user (such as basic attribute features, statistical attribute features, etc.) to construct an overall user portrait of the target user (this time it is represented as U).

[0061] On this basis, by executing the subsequent step S103, LLM can be used to analyze the learning needs of the target user (such as breaking down the user's intentions) to determine the calling of relevant intelligent agents to meet the real-time needs of the target user and the adjusted learning strategy.

[0062] S103: Utilize the user portrait of the target user, the second learning information of the target user, and the agent information, combined with the second prompt instruction prompt, and input them into the large language model to obtain the learning strategy for the target user output by the model.

[0063] In this embodiment, after obtaining the user portrait of the target user through step S102, and obtaining the first learning information such as the scenario information (such as review scenario, preview scenario, exam preparation scenario, holiday learning scenario, etc.) in which the target user is located and the second learning information such as the learning needs proposed by the target user (i.e., breaking down the specific intentions of the target user) through step S101, the user portrait of the target user, the learning needs proposed by the target user, the scenario information in which the target user is located, and the agent information can be further integrated into the prompt instruction prompt (defined here as the second prompt instruction) and input into the large language model (LLM) to obtain the personalized learning strategy for the target user output by the large language model, so as to execute the subsequent step S104.

[0064] Specifically, in order to make use of the explicit / implicit needs of the target users expressed by LLM, combined with user portraits and scenario information, to make decisions on personalized learning strategies, we first constructed a variety of different scenarios, different needs, and different implicit features, and combined the strategy decision data brought by different portraits to pre-train the LLM, so as to meet the actual needs of target users in real scenarios. In the process of training LLM, the descriptions and functions of different tools will be simultaneously trained in combination with prompts. In this way, the model can effectively decide which appropriate tool to apply when combined with specific scenarios and user intentions. The strategy decisions may include but are not limited to tool calls, learning plan output, evaluation feedback, etc.

[0065] In this way, when using LLM to generate learning strategies for target users, the user profiles (such as ∈U (including instant portrait + historical portrait)), learning needs proposed by target users ( ∈Q, can be omitted), the scene information of the target user ( ∈S) and agent information (represented as R in this case) are integrated into the prompt command as the input of LLM, so as to form a learning strategy for the target user at the tth moment through the natural language understanding ability and thinking logic reasoning ability of LLM itself (represented as ), the specific calculation formula is as follows:

[0066] Among them, the addition of agent information (R) and input parameter features and LLM output in the prompt instruction is to form a specific learning strategy. Specifically, LLM generates learning strategies based on the target user's profile, application scenario and real-time needs combined with the agent's input parameters and description. For example, assuming that the target user has a high fear of difficulty and frequently skips questions, the strategy generated by LLM after analysis may be: "Recommend objective questions with lower difficulty under weak points", and the sample content of R can be as follows: R:{ R1: {“Test paper assistant”: “Can build test papers and test question packages according to the user-specified scope and test question difficulty, etc.”, “Parameter input”: [grade, book_code, error_topic,...]} R2: {“Question-answering teacher”: “…”} }.

[0067] S104: Based on the learning strategy, the intelligent agent is collaboratively called through the large language model to achieve companion learning for the target user.

[0068] In this embodiment, after obtaining the learning strategy for the target user through step S103, further, each intelligent agent can be called through the large language model according to the learning strategy to realize the main process of companion learning for the target user, and in the process of realizing the detailed process of companion learning, the mutual calling between each intelligent agent can realize the collaborative calling of each intelligent agent, thereby completing the entire companion learning processing process for the target user.

[0069] For example: Figure 2 As shown in Figure 1, suppose that during a certain application practice, the target user (such as a student) said to the learning machine (deployed with the LLM mentioned above), "The final exam is coming soon. It would be great if you could help me with the special review" (such as Figure 2 In the query, at this time, LLM can quickly analyze the target user’s (such as students) intention as: to get a score before the exam. The key elements of this intention are as follows: Figure 2 The "bot: Final Exam, Score Improvement" shown in the figure can then call the generated relevant learning profile of the target user (such as a student) and match the weak learning resources corresponding to the target user (such as a student), and the learning benefits of these learning resources are calculated through the relevant learning benefit model, and the specific calculation process is not limited. Then, the call of the intelligent agent such as test paper composition and scoring is combined to complete the construction of the entire exercise package / volume. After the target user (such as a student) completes the study, the intelligent agent AI teacher and AI comment generation can be automatically called according to the error cause of the target user (such as a student), so as to achieve targeted learning and explanation of the questions that the target user (such as a student) does not know and plan subsequent practice plans. In this way, through an inadvertent word of feedback from the target user (such as a student), targeted practice before the exam and perfect guidance can be completed. Among them, the call of each of the aforementioned intelligent agents or tools is completely triggered by the LLM generation instruction parameters or some fixed workflow combination modes to ensure the accuracy of scheduling.

[0070] It should be noted that in the above scheduling process, LLM is mainly responsible for the connection of the main process, while each agent (such as Figure 2The "test paper setting", "grading", "AI teacher") in the system can be coordinated with each other according to the feedback from the environment, such as Figure 2 After the "AI teacher" has explained the test questions to the target user (such as students), he can dispatch the "test paper maker" agent to practice the same type of test paper. In this way, the main process LLM makes decisions based on user intentions and portraits, and the running processes in each agent in the detailed process can be coordinated and called, thereby improving decision-making efficiency.

[0071] On this basis, in order to further improve the learning companionship effect for the target user and further enhance the learning experience of the target user, an optional implementation method is to coordinately call each intelligent agent through LLM to realize the learning companionship process for the target user, and then use the process self-reflection mechanism to update the user portrait of the target user according to the feedback information of the target user, and then use the updated user portrait of the target user to update the learning strategy of the target user, and use the updated learning strategy of the target user to coordinately call the intelligent agent again through the large language model to realize intelligent learning companionship for the target user until the learning needs proposed by the target user are met. This effectively improves the learning companionship effect for the target user and further enhances the learning experience of the target user.

[0072] Specifically, in order to enable the companion learning system deployed with LLM to analyze and understand the learning situation of the target user (such as students) more like a human being and give them the correct learning strategy in time, LLM needs to be able to constantly reflect in the reasoning process and make adjustments in real time based on the individual portrait of the target user (such as students). For example, after the target user (such as student) receives the result output by the corresponding intelligent agent, what is the corresponding state? Is it timely and correctly fed back to the intelligent agent / LLM, etc., so as to enter a new round of learning strategy updates, realizing a process of real-time monitoring, self-learning and reasoning error correction.

[0073] In order to achieve the above process and improve the learning effect, this application proposes a "multi-layer self-reflective reasoning chain" based on LLM, such as Figure 3 As shown, the reasoning process of LLM is increased with comparison (comparison with the original portrait, intent, context, feedback facts, etc.) and depth of thinking. Specifically, it may include but is not limited to four key steps, namely: (1) correlation between recommended content and intent, (2) analysis of real content participation, analysis of learning process interaction & behavior, (3) analysis of learning results, and (4) updating (such as adding / modifying) user portraits. Among them, first, LLM needs to analyze and reason each step (step) in turn, and then reversely verify the results of each process in the next step. If it is reasonable, the last step of updating (such as adding / modifying) the user portrait is performed. If the reverse reasoning is unreasonable, it is necessary to further trigger multiple rounds of interactive confirmation or temporarily shelve it for verification. After the target user's portrait is updated, a new learning strategy is triggered ( ), ensuring the real-time learning effect, thereby greatly promoting the correctness of the analysis process and reducing the illusion of LLM-generated content.

[0074] Here, it should be noted that the general LLM base itself may not be able to complete the above step analysis well according to the prompt, so this application constructs the above-mentioned thinking reasoning chain data alone, and then performs supervised fine-tuning training on the LLM. Moreover, unlike the traditional LLM decoding process which only outputs the topN best results, in the LLM application (reasoning) process, this application requires the LLM to be able to generate multiple results, and then score the steps through the reward-model (scoring model) trained by the supervised task, so as to screen each step in the decoding process, and the judgment of the best step comes from the feedback of the end user's learning outcomes.

[0075] On this basis, an optional implementation method is to update the user profile of the target user according to the feedback information of the target user and utilize the process self-reflection mechanism during the learning process. The implementation process may specifically include the following steps AD: Step A: During the learning process, determine the target learning resources to be recommended based on the learning strategy; and determine whether the relevance of the target learning resources to the target user's learning needs and user profile meets the preset requirements.

[0076] In this implementation, after LLM outputs the learning strategy for the target user, it can further determine the learning resources to be recommended (this time it is defined as the target learning resource) according to the learning strategy; and judge whether the relevance between the target learning resource and the learning needs and user profile of the target user meets the preset requirements (the specific content is not limited and can be set according to the actual situation and experience value), such as Figure 3 As shown, to confirm self-consistency, if yes, proceed to the subsequent step B; if not, proceed to the subsequent step C.

[0077] Step B: If so, the target learning resources are recommended to the target user, and real-time interaction is performed with the target user based on the target user's learning participation, interaction during participation, and various learning behaviors and learning attitudes, so as to recommend the adjusted target learning resources to the target user and update the relevant content in the target user's user profile based on the target user's feedback information.

[0078] If it is determined through step A that the relevance of the target learning resource to the target user's learning needs and user profile meets the preset requirements (i.e., the content of the target learning resource meets the target user's learning intention), the target learning resource can be further recommended (output) to the target user, such as Figure 3As shown, the learning participation of the target user is monitored, and the interaction and various learning behaviors and learning attitudes of the target user during the participation process are observed, such as low-frequency participation, exit from the learning process, poor performance during the process (such as expression of aversion to learning), etc., and real-time interaction is carried out with the target user based on this information, so as to further evaluate whether the content quality of the target learning resource meets the needs of the target user, whether the content is too difficult, the amount of content, and changes in the student's learning status, etc., based on the feedback information of the target user, so as to adjust the learning strategy and the target learning resources it contains, and recommend the adjusted target learning resources to the target user to perform the subsequent step D. For example, if the target user (such as a student) encounters a difficult problem and makes no progress, the relevant AI teacher can be called for auxiliary learning; or, if the target user (such as a student) is in a bad mood for learning, the relevant mood relief agent can be called for auxiliary learning. At the same time, the relevant content in the user portrait of the target user can also be updated.

[0079] Step C: If not, confirm the adjusted target learning resources through secondary interaction with the target user; recommend the adjusted target learning resources to the target user, and update the relevant content in the user profile of the target user.

[0080] If it is determined through step A that the relevance between the target learning resource and the target user's learning needs and user profile does not meet the preset requirements (i.e., the content of the target learning resource does not meet the target user's learning intention), then this recommendation can be abandoned, and a second interaction can be conducted with the target user to confirm the adjusted target learning resource, and the adjusted target learning resource can be recommended to the target user for subsequent step D. At the same time, the relevant content in the target user's user profile can also be updated.

[0081] Step D: Analyze the learning results of the target user for the target learning resource and / or the adjusted target learning resource, and update the user profile of the target user based on the analysis results.

[0082] After recommending the target learning resources and / or adjusted target learning resources to the target user through step B or C, the target user's learning results for the target learning resources and / or adjusted target learning resources can be further analyzed, and the user profile of the target user can be updated based on the analysis results.

[0083] Among them, this application does not limit the specific content of the analysis results, which may include but are not limited to the target user's insufficient completion of the target learning resources and / or the learning content of the target learning resources and the analysis results of the causes of wrong questions.

[0084] Specifically, after the target users have completed learning the target learning resources and / or adjusted target learning resources, the learning outcomes are evaluated, such as insufficient completion of the learning content, analysis of the causes of wrong questions, etc., and interactive communication is conducted with the target users to update the target users' (such as students') weaknesses and other portrait elements.

[0085] In this way, through the above reflection process, the accuracy of recommendation of learning resources for target users in the learning process can be improved and problems can be discovered in time to form reliable responses, thereby improving the learning experience of target users.

[0086] For example: Assume that the query raised by the target user (such as a student) is: "Help me review the recent courses."

[0087] The contents of the target learning resources are: [“Explanation of the concept of quadratic equations.mp4”, “Comprehensive application of quadratic equations.mp4”, “Concept exercises of quadratic equations.txt”].

[0088] User portrait of target users (such as students): [students who are still learning, fear of difficulty, practice multiple types of questions, lack of sense of learning goals, current learning progress, textbook version, historical weaknesses...].

[0089] LLM first analyzes the recent courses of the target user (such as students), and frames it as the "Quadratic Equations" chapter through the learning progress and textbook version in the portrait, but is unable to lock in the specific section, triggering the interaction between the quadratic user and the target user (such as students): "Can you tell me in detail which section of the quadratic equations it is?"

[0090] The query raised by the target user (such as students) may be: "Basic concepts".

[0091] The adjusted target learning resource launched by LLM can be: "Explanation of the concept of quadratic equations.mp4".

[0092] However, the target users (such as students) quickly skipped the learning process. The abnormal behavior triggered the thinking about the profile. LLM found that the target users (such as students) were the type who practiced questions. It further recommended the adjusted target learning resources as: "Concept exercises on quadratic equations" and recommended "Concept exercises on quadratic equations" in a gradually increasing difficulty level from medium difficulty.

[0093] At the same time, the target user's (such as a student's) answer results are used to update the target user's (such as a student's) portrait, whether the "quadratic equation concept" is a persistent weak point, the time spent in the learning process and the emotion analysis of the learning process, in addition to the target user's (such as a student's) current learning status and recommendation satisfaction in the portrait, to optimize the decision of the learning strategy.

[0094] In this way, by executing the above steps S101-S104, for the first time, based on LLM, the unstructured information that is difficult to be processed by other traditional small models such as pre-trained language representation models in the multi-dimensional features such as real-time dialogue, behavior, expression, feedback, etc. of the target user is structured, and the personalized user portrait of the target user is constructed according to the full-scene multi-modal learning features contained in the processing results, and then the learning intention of the target user is disassembled in real time through single / multi-round interaction, behavior process, expression feedback, etc., and the decision of the learning strategy is obtained according to the role required by the companion learning, so as to guide multiple intelligent agents to call and coordinate multiple types of auxiliary tools, and build a multi-agent collaborative companion learning system with LLM as the core decision-making brain to fully analyze the learning demands, learning status and feelings of the target users (such as students) throughout the whole process, and call and adjust the learning strategy in time. In particular, in order to improve the efficiency of companion learning and achieve a human-like companion effect, this application introduces a process reflection mechanism in the companion learning process to promote the accurate construction of the target user portrait by LLM, thereby more efficiently improving the timeliness and accuracy of the entire companion learning process. It realizes multifunctional intelligent learning companionship, which not only improves the education quality for target users, but also increases the fun of target users in the learning process.

[0095] In summary, the present embodiment provides an intelligent learning companion method, which first obtains the first learning information and the second learning information of the target user; then uses the first learning information of the target user, combined with the first prompt instruction prompt, and inputs it into the LLM to obtain the user portrait of the target user output by the model; then uses the user portrait of the target user, the second learning information of the target user and the agent information, combined with the second prompt instruction prompt, and inputs them into the LLM to obtain the learning strategy for the target user output by the model; and then, according to the learning strategy, collaboratively calls the agent through the LLM to realize learning companionship for the target user.

[0096] It can be seen that since the present application first utilizes the target user's first learning information (such as the target user's historical learning information, current learning status, and scene information) based on LLM, a full-scene portrait of the target user is accurately constructed, and then the learning strategy is determined based on the full-scene portrait of the target user and the target user's second learning information (such as the learning needs proposed by the target user) and the intelligent agent information. Based on the learning strategy, through the collaborative calling of the intelligent agent combined with the process self-reflection mechanism, intelligent learning companionship for the target user is realized, which effectively improves the timeliness and accuracy of the entire learning companionship process, thereby improving the learning companionship effect for the target user, and then improving the learning experience of the target user.

[0097] Second embodiment

[0098] This embodiment will introduce an intelligent learning companion device. For related content, please refer to the above method embodiment.

[0099] See also Figure 4 , is a schematic diagram of the composition of an intelligent learning companion device provided in this embodiment, the device 400 includes: An acquisition unit 401 is used to acquire first learning information and second learning information of a target user; The first input unit 402 is used to use the first learning information of the target user, combined with the first prompt instruction prompt, to input into the large language model to obtain the user portrait of the target user output by the model; The second input unit 403 is used to use the user portrait of the target user, the second learning information of the target user and the agent information, combined with the second prompt instruction prompt, to input into the large language model to obtain the learning strategy for the target user output by the model; The companion learning unit 404 is used to collaboratively call the intelligent agent through the large language model according to the learning strategy to achieve companion learning for the target user.

[0100] In an implementation of this embodiment, the first learning information of the target user includes the current learning state of the target user, the historical learning information of the target user, and the scene information of the target user; the first input unit 402 includes: The first input subunit is used to perform structured processing on the historical learning information of the target user, and use the obtained processing result in combination with the first prompt instruction prompt to input it into the large language model to obtain the historical portrait of the target user output by the model; The second input subunit is used to utilize the current learning state of the target user, the historical portrait of the target user, and the scene information of the target user, combined with the updated first prompt instruction prompt, and input them into the large language model to obtain the instant portrait of the target user output by the model; The construction subunit is used to construct an overall user profile of the target user by using the historical profile of the target user and the instant profile of the target user.

[0101] In one implementation of this embodiment, the second learning information of the target user includes the learning needs proposed by the target user; the learning needs proposed by the target user are determined based on at least one round of conversation content between the large language model and the target user.

[0102] In an implementation of this embodiment, the second input unit 403 is specifically used to: The user portrait of the target user, the learning needs proposed by the target user, the scene information of the target user and the agent information are used, combined with the second prompt instruction prompt, and input into the large language model to obtain the learning strategy for the target user output by the model.

[0103] In one implementation of this embodiment, the companion learning unit 404 is specifically used for: According to the learning strategy, each intelligent agent is called through the large language model to realize the main process of companion learning for the target user; and in the process of realizing the detailed process of companion learning, the mutual calling between each intelligent agent is realized to realize the coordinated calling of each intelligent agent, so as to complete the whole companion learning processing process for the target user.

[0104] In an implementation of this embodiment, the device further includes: An updating unit, used to update the user profile of the target user by using a process self-reflection mechanism according to the feedback information of the target user during the companion learning process; The calling unit is used to use the updated user portrait of the target user to update the learning strategy of the target user, and use the updated learning strategy to collaboratively call the intelligent agent again through the large language model to achieve intelligent learning companionship for the target user until the learning needs proposed by the target user are met.

[0105] In an implementation of this embodiment, the updating unit includes: A judgment subunit is used to determine the target learning resources to be recommended according to the learning strategy during the companion learning process; and to judge whether the relevance between the target learning resources and the learning needs and user profile of the target user meets the preset requirements; A first updating subunit is used for recommending the target learning resource to the target user if it is determined that the relevance between the target learning resource and the learning needs and user profile of the target user meets preset requirements, and for interacting with the target user in real time according to the learning participation, interaction during the participation process, and various learning behaviors and learning attitudes of the target user, so as to recommend the adjusted target learning resource to the target user according to the feedback information of the target user, and to update the relevant content in the user profile of the target user; The second updating subunit is used to confirm the adjusted target learning resource through secondary interaction with the target user if it is determined that the relevance between the target learning resource and the learning needs and user profile of the target user does not meet the preset requirements; and recommend the adjusted target learning resource to the target user, and update the relevant content in the user profile of the target user; The third updating subunit is used to analyze the learning results of the target user on the target learning resource and / or the adjusted target learning resource, and update the user portrait of the target user according to the analysis results.

[0106] In one implementation of this embodiment, the analysis result includes an analysis result of the target user's insufficient completion of the target learning resource and / or the learning content of the target learning resource and the cause of the wrong questions.

[0107] In one implementation of this embodiment, the current learning status of the target user includes at least one of the current expression of the target user, the text proposed and / or the language and behavioral expressions issued by the target user; the historical learning information of the target user includes at least one of the historical behavior sequence, interaction sequence, answer content, process performance and results of the target user; the scene information of the target user includes at least one of the review scene, preview scene, exam preparation scene, and holiday learning scene.

[0108] Furthermore, the embodiment of the present application also provides an intelligent learning companion device, including: a processor, a memory, and a system bus; The processor and the memory are connected via the system bus; The memory is used to store one or more programs, and the one or more programs include instructions. When the instructions are executed by the processor, the processor executes any one of the implementation methods of the above-mentioned intelligent companion learning method.

[0109] Furthermore, an embodiment of the present application also provides a computer-readable storage medium, in which instructions are stored. When the instructions are executed on a terminal device, the terminal device executes any one of the implementation methods of the above-mentioned intelligent companion learning method.

[0110] Furthermore, an embodiment of the present application also provides a computer program product, which, when executed on a terminal device, enables the terminal device to execute any one of the implementation methods of the above-mentioned intelligent companion learning method.

[0111] It can be known from the description of the above implementation mode that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment method can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product can be stored in a storage medium such as ROM / RAM, a disk, an optical disk, etc., including several instructions for enabling a computer device (which can be a personal computer, a server, or a network communication device such as a media gateway, etc.) to execute the methods described in the various embodiments of the present application or certain parts of the embodiments.

[0112] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0113] It should also be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0114] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An intelligent learning companion method, characterized in that: include: Acquire first learning information and second learning information of a target user; Using the first learning information of the target user, combined with the first prompt instruction prompt, inputting it into the large language model, and obtaining a user portrait of the target user output by the model; Utilizing the user portrait of the target user, the second learning information of the target user, and the agent information, combined with the second prompt instruction prompt, inputting them into the large language model, and obtaining a learning strategy for the target user output by the model; According to the learning strategy, the agent is collaboratively called through the large language model to achieve companion learning for the target user.

2. The method according to claim 1, characterized in that The first learning information of the target user includes the current learning state of the target user, the historical learning information of the target user, and the scene information of the target user; the first learning information of the target user is used in combination with the first prompt instruction prompt, and is input into the large language model to obtain the user portrait of the target user output by the model, including: Structuring the historical learning information of the target user, and using the obtained processing result in combination with the first prompt instruction prompt, inputting it into the large language model to obtain a historical portrait of the target user output by the model; Using the current learning state of the target user, the historical portrait of the target user, and the scene information of the target user, combined with the updated first prompt instruction prompt, input them into the large language model to obtain the instant portrait of the target user output by the model; The overall user profile of the target user is constructed by using the historical profile of the target user and the instant profile of the target user.

3. The method according to claim 1, characterized in that The second learning information of the target user includes the learning needs proposed by the target user; the learning needs proposed by the target user are determined based on at least one round of conversation content between the large language model and the target user.

4. The method according to claim 3, characterized in that The user portrait of the target user, the second learning information of the target user and the agent information are combined with the second prompt instruction prompt, input into the large language model, and the learning strategy for the target user output by the model is obtained, including: The user portrait of the target user, the learning needs proposed by the target user, the scene information of the target user and the agent information are used, combined with the second prompt instruction prompt, and input into the large language model to obtain the learning strategy for the target user output by the model.

5. The method according to claim 1, characterized in that The step of collaboratively calling the agent through the large language model according to the learning strategy to achieve companion learning for the target user includes: According to the learning strategy, each intelligent agent is called through the large language model to realize the main process of companion learning for the target user; and in the process of realizing the detailed process of companion learning, the mutual calling between each intelligent agent is realized to realize the collaborative calling of each intelligent agent, and the whole companion learning processing process for the target user is completed.

6. The method according to any one of claims 1 to 5, characterized in that: The method further comprises: In the learning process, based on the feedback information of the target user, the user profile of the target user is updated by using the process self-reflection mechanism; The updated user portrait of the target user is used to update the learning strategy of the target user, and the updated learning strategy is used to collaboratively call the intelligent agent again through the large language model to achieve intelligent learning companionship for the target user until the learning needs proposed by the target user are met.

7. The method according to claim 6, characterized in that In the accompanying learning process, according to the feedback information of the target user, the user profile of the target user is updated by using the process self-reflection mechanism, including: In the learning process, the target learning resources to be recommended are determined according to the learning strategy; and it is determined whether the relevance between the target learning resources and the learning needs and user profile of the target user meets the preset requirements; If so, the target learning resource is recommended to the target user, and real-time interaction is performed with the target user based on the target user's learning participation, interaction during participation, and various learning behaviors and learning attitudes, so as to recommend the adjusted target learning resource to the target user and update the relevant content in the user profile of the target user based on the target user's feedback information; If not, confirm the adjusted target learning resource through secondary interaction with the target user; recommend the adjusted target learning resource to the target user, and update the relevant content in the user profile of the target user; The learning results of the target user on the target learning resource and / or the adjusted target learning resource are analyzed, and the user portrait of the target user is updated according to the analysis results.

8. The method according to claim 7, characterized in that The analysis result includes the target user's insufficient completion of the target learning resource and / or the learning content of the target learning resource and the analysis result of the reasons for the wrong questions.

9. The method according to claim 2, characterized in that: The current learning state of the target user includes at least one of the current expression of the target user, the text proposed by the target user and / or the language and behavior expression issued by the target user; The historical learning information of the target user includes at least one of the target user's historical behavior sequence, interaction sequence, answer content, process performance and results; the scenario information of the target user includes at least one of the review scenario, preview scenario, exam preparation scenario and holiday learning scenario.

10. An intelligent learning companion device, characterized in that: include: An acquisition unit, used for acquiring first learning information and second learning information of a target user; A first input unit is used to use the first learning information of the target user, combined with a first prompt instruction prompt, to input into the large language model to obtain a user portrait of the target user output by the model; A second input unit is used to use the user portrait of the target user, the second learning information of the target user and the agent information, combined with the second prompt instruction prompt, to input into the large language model to obtain the learning strategy for the target user output by the model; A companion learning unit is used to collaboratively call the intelligent agent through the large language model according to the learning strategy to achieve companion learning for the target user.

11. An intelligent learning companion device, characterized in that: include: Processor, memory, system bus; The processor and the memory are connected via the system bus; The memory is used to store one or more programs, wherein the one or more programs include instructions, and when the instructions are executed by the processor, the processor executes the method according to any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, and when the instructions are executed on a terminal device, the terminal device executes the method according to any one of claims 1 to 9.

Images