Learning resource pushing method and application based on interactive language model

Abstract

The invention discloses a learning resource pushing method based on an interactive language model, and the method comprises the steps: obtaining learning interaction data of a learner, carrying out the semantic analysis of the learning interaction data based on a large language model, and extracting key features; performing feature enhancement on the answer text information and the knowledge point text information of the key features, constructing a knowledge state diagnosis model based on the enhanced features, and predicting the knowledge state of the learner through a forgetting attenuation coefficient matrix of the knowledge state diagnosis model; and performing feature alignment and mutual matching on the learner knowledge state and the knowledge point label of the learning interaction data, and extracting learning resource text content according to a matching result and pushing the learning resource text content. The problem that the state of a learner cannot be accurately evaluated in real time and learning resources cannot be individually matched in a traditional education mode can be solved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence and machine learning technology, and more specifically, to a learning resource push method, device, electronic device, and computer-readable storage medium based on an interactive language model. Background Art

[0002] In the field of educational technology, with the development of artificial intelligence and machine learning, personalized learning has gradually become a key direction of educational reform. The core of personalized learning lies in providing customized learning content and learning paths based on learners' individual characteristics, knowledge level, and learning needs, in order to improve learners' learning efficiency and quality.

[0003] However, achieving this goal faces two major challenges: (1) Accuracy and timeliness of learning status assessment: Traditional learning status assessment methods rely on teachers' subjective judgment and regular testing, which is not only highly subjective but also difficult to achieve real-time monitoring and timely feedback. (2) Personalized matching of learning resources: Among the vast amount of learning resources, how to quickly and accurately match the most appropriate learning resources based on the learners' specific needs and knowledge status is a technical challenge.

[0004] With the rapid development of educational technology, the demand for personalized learning is growing. However, traditional learning status assessment methods and learning resource matching technologies are no longer able to meet the needs of real-time, accurate personalized learning. Therefore, there is an urgent need for a new method that can monitor learners' knowledge status in real time, accurately construct knowledge point labels, and effectively match personalized learning resources to improve learning efficiency and quality, and promote learners' knowledge acquisition and ability development. Summary of the Invention

[0005] In response to at least one defect or improvement need in the prior art, the present invention provides a learning resource push method and application based on an interactive language model, aiming to solve the problem in the traditional education model that it is impossible to accurately evaluate the learner's status in real time and match learning resources in a personalized manner.

[0006] To achieve the above-mentioned objectives, according to the first aspect of the present invention, a method for pushing learning resources based on an interactive language model is provided, comprising: obtaining the learner's learning interaction data, performing semantic analysis on the learning interaction data based on a large language model, and extracting key features; performing feature enhancement on the answer text information and knowledge point text information of the key features, constructing a knowledge state diagnosis model based on the enhanced features, and predicting the learner's knowledge state through the forgetting attenuation coefficient matrix of the knowledge state diagnosis model; aligning the learner's knowledge state with the knowledge point labels of the learning interaction data and matching them with each other, extracting the learning resource text content according to the matching results and pushing them.

[0007] In one embodiment of the present invention, the step of obtaining the learner's learning interaction data includes: obtaining the learner's interactive answer record set based on the database of the online learning platform. , where the learner's answer record set It includes specific information of each answer of the learner, including question information, knowledge point information, answer time, and correctness of the answer; extracts learning resources available in the platform, including teaching videos, course materials, and exercises related to course knowledge points, extracts text content from the learning resources, and saves them as corresponding text files; each learner's answer record and the learning resources collected by the platform are divided into different files for storage according to teaching video resources, course materials, and exercises related to course knowledge points, where each file contains the corresponding text file.

[0008] In one embodiment of the present invention, after obtaining the learner's learning interaction data, it also includes: deleting and completing the answer sequence in the learning interaction data, and sorting it by timestamp; encoding the learning interaction data using UTF-8, standardizing it into text data, performing word segmentation on the text data based on the Tokenizer, and using the LLM module to generate feature embedding; dividing the learning interaction data into several equal parts, part of which is used for model training, part for model verification, and part for model testing.

[0009] In one embodiment of the present invention, the knowledge state diagnosis model includes a first sub-network model and a second sub-network model. The first sub-network model is used to construct a complete embedded expression of the learner's answer information in combination with the text features enhanced by the large language model features, and output a potential feature vector; the second sub-network model is used to perform a decoding operation on the acquired embedded expression and obtain the final prediction result in combination with the forgetting decay coefficient matrix.

[0010] In one embodiment of the present invention, the execution steps of the first sub-network model include: using the pre-trained large language model to extract features of the learner's answer text and knowledge point text information; obtaining a complete embedded expression based on the feature extraction, and combining the key features extracted by the large language model with other relevant features; constructing a positional encoding based on the self-attention mechanism of Transformer, sending the complete embedded expression to the encoder, capturing the complex logical relationship between knowledge points and exercises, and thus outputting a potential feature vector.

[0011] In one embodiment of the present invention, the execution steps of the second sub-network model include: performing a decoding operation based on the potential feature vector, wherein the potential features of the knowledge point are used as a query function to obtain the learner's knowledge mastery of different knowledge points; constructing a forgetting decay coefficient matrix based on the time distance between the questions, which is used to adjust the importance of each time step in the output sequence, and multiplying the output sequence by the forgetting decay coefficient matrix to adjust the output sequence; outputting the predicted probability of knowledge mastery based on the output sequence, inputting the potential feature vector into the fully connected layer for dimensional feature change, and converting it into a target probability value in combination with the activation function.

[0012] In one embodiment of the present invention, the learning resource push step includes: using measurement methods such as cosine similarity to align the learner's mastery of each knowledge point with the knowledge point label of the learning resource, and identifying the learning resource that best matches the learner's current knowledge status based on the alignment result; considering the release time and preset score of the learning resource to calculate the weight, and normalizing the weights of all resources to ensure that the total weight is 1; selecting the learning resource with the highest weight for push based on the normalized weight and matching result.

[0013] According to the second aspect of the present invention, a learning resource pushing device based on an interactive language model is also provided, which includes: a key feature extraction module for obtaining learners' learning interaction data, performing semantic analysis on the learning interaction data based on a large language model, and extracting key features; a knowledge state prediction module for performing feature enhancement on the answer text information and knowledge point text information of the key features, constructing a knowledge state diagnosis model based on the enhanced features, and predicting the learner's knowledge state through the forgetting attenuation coefficient matrix of the knowledge state diagnosis model; a learning resource pushing module for aligning and matching the learner's knowledge state with the knowledge point labels of the learning interaction data, extracting the learning resource text content according to the matching results, and pushing it.

[0014] According to the third aspect of the present invention, an electronic device is also provided, which includes at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program, and when the computer program is executed by the processing unit, the processing unit performs the steps of the method described in any one of the above embodiments.

[0015] According to a fourth aspect of the present invention, a computer-readable storage medium is also provided, which stores a computer program executable by an access authentication device. When the computer program runs on the access authentication device, the access authentication device executes the steps of the method described in any one of the above embodiments.

[0016] In general, the above technical solutions conceived by the present invention can achieve at least the following beneficial effects compared with the prior art: 1) By combining the Large Language Model (LLM) with learner knowledge state diagnosis technology, this invention can more accurately extract the learner's knowledge state. The feature enhancement part integrates text semantic features and metadata features, and uses the self-attention mechanism and forgetting decay coefficient matrix to capture complex temporal dynamics and knowledge point associations, thereby achieving high-precision modeling of the learner's knowledge mastery level. 2) We built a feature selection and alignment module (FSM) to accurately match learners' knowledge mastery status with the knowledge point labels of learning resources. We also introduced a weighted calculation based on resource release time and ratings to further optimize the ranking of recommended resources. This approach, combined with multi-dimensional weight distribution, significantly improved the pertinence and practicality of resource recommendations, providing learners with the learning resources that best meet their current needs. 3) Through end-to-end automated processing, a complete process is formed, from data collection and cleaning to feature extraction, status prediction, and resource matching and delivery. The system dynamically adjusts recommendation strategies based on learners' real-time interactions, offering strong adaptability and continuously optimizing the delivery of learning resources. This combination of automation and dynamic adjustments significantly enhances the intelligence of the learning platform and the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 A diagram illustrating the execution steps of a learning resource push method based on an interactive language model provided in an embodiment of the present application; Figure 2 A schematic diagram of data acquisition in an online learning environment provided in an embodiment of the present application; Figure 3 A network diagram of learner status perception and resource push provided in an embodiment of the present application; Figure 4 A schematic diagram of the structure of a learning resource push device based on an interactive language model provided in an embodiment of the present application; Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application; Figure 6A schematic diagram of the structure of a computer-readable storage medium provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0020] The terms "first," "second," "third," and the like in the specification and claims of this application and the accompanying drawings are used to distinguish between different objects, not to describe a particular order. Furthermore, the terms "including," "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or elements is not limited to the listed steps or elements, but may optionally include steps or elements not listed, or may optionally include other steps or elements inherent to the process, method, product, or apparatus.

[0021] The first embodiment of the present invention proposes a learning resource push method based on an interactive language model, for example, including: step S1, obtaining the learner's learning interaction data, performing semantic analysis on the learning interaction data based on a large language model, and extracting key features; step S2, performing feature enhancement on the answer text information and knowledge point text information of the key features, constructing a knowledge status diagnosis model based on the enhanced features, and predicting the learner's knowledge status through the forgetting attenuation coefficient matrix of the knowledge status diagnosis model; step S3, aligning the learner's knowledge status with the knowledge point labels of the learning interaction data and matching them with each other, extracting the learning resource text content according to the matching results and pushing them.

[0022] like Figure 1 As shown, in step S1, for example, based on the learner's online learning environment, the learner's answer interaction data on the junior high school history subject on the online education platform is collected, including answer records, learning time, etc., and at the same time, relevant information of the learning resources is obtained and the text content of the learning resources is formatted.

[0023] like Figure 2 As shown, learners are studying related courses through the junior high school history online learning platform. In this scenario, the learners' answer records will be collected and saved, and various learning resources in the platform will also be recorded and saved in a unified format.

[0024] Based on the database of the junior high school history online learning platform, the learner Xiaolin's answer record collection for the history subject is obtained , the question types are all multiple choice questions, among which Xiaolin's answer record collection Includes specific information about each answer, such as question information , knowledge point information , answering time , answer the situation , the specific format of an interaction record is as follows: { "student_id":2024011328; "student_name":Xiao Lin; "question_id":1382721384; "question_text":"Did the Three Kingdoms period begin during the Eastern Han Dynasty or the Western Han Dynasty?"; "concept_id":76134241122343; "concept_text":"History of the Three Kingdoms"; "answer_time":"2024-10-09-22:34:42"; “answer”:1 } Extract the learning resources about junior high school history available on the platform and their corresponding ratings and release times. Learning resources mainly include history teaching videos. , course materials , Exercises related to course knowledge points , extract the text content of these three learning resources and save them as corresponding JSON files. Among them, the JSON file is in dictionary format, with the key being the corresponding learning resource id and the value being the corresponding text content.

[0025] The answer records of each learner and the learning resources collected by the platform are stored in files. The learner's answer records are named with "learner ID". For example, the learning resources of "Xiaolin-2023093873" are stored in three different folders. 、 、 Store teaching video resources separately , course materials and exercises related to course knowledge points , where each folder contains its corresponding text JSON file.

[0026] Furthermore, the collected data is cleaned and preprocessed, and semantic analysis of knowledge points, exercises, and learning resource tags is performed based on the large language model (LLM). Key features are extracted, and embedded representations are constructed for each, mapped to a high-dimensional vector space, and feature alignment is achieved to facilitate subsequent matching and recommendation.

[0027] Specifically, for learners' interactive data on answering questions Perform preprocessing operations, delete the current row of interactive data for the default value in the data, and n Data with less than 5 times will be deleted. n The data with more than 100 times is split and completed, and the learner's answer data is sorted by timestamp: . Preprocessing operations are performed on the text content under the learning resources. First, the text data is standardized and UTF-8 is uniformly used for text encoding. The text is segmented based on the Tokenizer to facilitate the next step of LLM to generate feature embedding.

[0028] Taking data distribution into consideration, for example, the pre-processed respondent interaction data is reasonably divided into 10 equal parts based on the learner's age and answering time distribution, of which 8 parts are used for model training, 1 part is used for model verification, and 1 part is used for model testing, that is, 80% of the data is used as a training set, 10% of the data is used as a verification set, and 10% of the data is used as a test set.

[0029] In step S2, for the existing learner answer sequence, the feature enhancement of the exercise text information and the knowledge point text information is performed based on the pre-trained Llama-3-8B large language model, which is used as the feature input of the network model LLM2KT, and combined with the forgetting decay coefficient matrix to achieve the final prediction, output the learner's mastery probability on the specific knowledge point, and obtain the learner's knowledge mastery status. The LLM2KT network model includes the encoder MTEncoder and the decoder DFDecoder.

[0030] like Figure 3 As shown in the figure, in this example, the learner knowledge state perception model LLM2KT includes an encoder MTEncoder, a feature enhancement module, and a decoder DFDecoder module. The specific model training process consists of the following steps: The encoder MTEncoder is used to combine the text features enhanced by LLM features to construct a complete embedding expression of the learner's answer information , and output the latent feature vector.

[0031] Specifically, it includes feature enhancement: based on the trained Llama3-8B fine-tuning model, feature extraction is performed on the text content of the learner's answer and the text information of the knowledge points: ; The large language model is based on the pre-trained Llama-3-8B model. and They are respectively the question text information and the knowledge point text information, and Enhance embedding representations for learned features.

[0032] Feature embedding construction: Obtain a complete embedding expression, combining the features extracted by LLM with other available metadata features such as answer time , the ID information of the topic And the ID information of the knowledge point Combined: , in, For feature concatenation, in order to achieve dimensionality unification, we set the learned feature representation dimensions to 128, where the feature enhancement embedding representation and Dimensionality reduction is required.

[0033] Self-attention output: Based on the Transformer's self-attention mechanism, position encoding is constructed based on the learner's interactive position information. : ; in, represents the positional encoding that changes with the learner's interaction sequence position. Secondly, the complete embedding expression This is fed into the Encoder, which uses a multi-head self-attention mechanism to capture the complex logical relationship between knowledge points and exercises: ; ; ; in, It is a scaling factor used to control the dimension change. By splicing the weight coefficients obtained from different attention heads, the potential feature vector is output. : ; ; Among them, MLP represents the fully connected operation and LayerNorm represents layer normalization.

[0034] based on Figure 3 The decoder DFDecoder in the previous module is used to express the embedded Perform decoding operation and combine with forgetting decay coefficient matrix to obtain the final prediction result , including the following steps: Based on the potential feature vector of the previous module, a decoding operation is performed, in which the potential features of the knowledge point are used as the query, and the potential feature vector of the previous module is used as the key and value to obtain the output of each attention head: ; ; The weight coefficients of each attention head are concatenated, and combined with layer normalization and a fully connected layer, the learner's specific knowledge mastery level at different knowledge points is output: ; ; The learner's forgetting benefit gradually increases with time. In order to take into account the learner's forgetting benefit, the forgetting characteristics are modeled based on the time interval between different historical questions. The modeled forgetting decay coefficient matrix can be expressed as ,in i and j Represents the interval between two times: ; in, It is an artificially fixed coefficient used to represent the impact of time efficiency. The matrix is ​​used to adjust the importance of each time step in the output sequence. The output sequence is multiplied by the forgetting decay coefficient matrix to adjust the output sequence.

[0035] Combined with the forgetting effect, the interactive output sequence is used to obtain the predicted probability This step mainly inputs the potential vector into the fully connected layer to change the dimensional features, and combines it with the Softmax function to convert it into the final probability value : ; in, is the learnable weight coefficient, is the bias matrix, is the sigmoid activation function.

[0036] In order to train a more accurate learner state perception model and minimize the error between the predicted probability and the learner's actual answer, the model uses the binary cross entropy function as the model's loss function for training: ; in, and They represent the correctness of the learner's answer and the predicted output result of the model respectively. At the same time, the Adam optimizer is used to train the parameters of the entire model based on the gradient descent optimization algorithm of the present invention. The learning rate is set to 0.001. After 30 epochs, the model training is realized. We measure the prediction accuracy of the model by the AUC value, and take the best performance in the training process as the final prediction model.

[0037] In step S3, a feature selection module (FSM) is constructed to align and match the learner's knowledge status with the knowledge point labels of the learning resources. The weight distribution is then combined with the release time and rating of the learning resources to achieve the final learning resource push.

[0038] like Figure 3 As shown in the model diagram, this paper constructs a potential feature label representation for junior high school history learning resources based on the LLM large language model. At the same time, the ratings and release time of the learning resources also play a role in the final recommendation coefficient. The specific matching and push steps are as follows: Feature alignment and feature matching: Use cosine similarity to measure learners’ mastery of each knowledge point Knowledge point tags with learning resources Perform alignment and, based on the alignment results, identify the learning resources that best match the learner's current knowledge state; ; in, Represents the cosine similarity calculation. The numerator is the inner product of the feature vector of the learner's mastery of a specific knowledge point and the potential label vector of the learning resource. The denominator is the product of the two vector moduli. The final cosine similarity value is between -1 and 1. A value of 1 indicates that the two vectors are exactly the same, that is, the similarity is the highest. A value of 0 indicates that the two vectors are orthogonal, that is, completely dissimilar. A value of -1 indicates that the two vectors are completely opposite.

[0039] Weight distribution: Consider the weight calculation of release time and score, combined with the release time of learning resources and ratings To calculate the weight of each resource, normalize the weights of all resources to ensure that the total weight is 1; ; in, and The value range is between 0 and 1, and the value is manually specified. Reflecting the impact of time efficiency, Used to reflect the impact of resource scoring.

[0040] Weighted matching and learning resource push: Based on the normalized weights and matching results, the learning resource with the highest weight is selected for push: ; in, To obtain the maximum value, is a probability value list of learning resources. For the number of pushed learning resources of different types, the present invention sets the number of video resources to 2, the number of text resources to 3, and the number of exercises to 10. Therefore, the final number of pushed learning resources is 15, including three different types of learning resources:

[0041] In this example, based on the learner Xiaolin's potential knowledge, it is concluded that his knowledge of the Three Kingdoms period is not very high. The final learning resource pushed is as follows, where the exercises will be presented in the form of a set of exam papers, so they are presented as a single link:

[0042] In addition, if Figure 4 As shown, the second embodiment of the present invention proposes a learning resource pushing device 20 based on an interactive language model, for example, including: a key feature extraction module 201, a knowledge state prediction module 202 and a learning resource pushing module 203.

[0043] Among them, the key feature extraction module 201 is used to obtain the learner's learning interaction data, perform semantic analysis on the learning interaction data based on the large language model, and extract key features; the knowledge status prediction module 202 is used to enhance the features of the answer text information and knowledge point text information of the key features, build a knowledge status diagnosis model based on the enhanced features, and predict the learner's knowledge status through the forgetting attenuation coefficient matrix of the knowledge status diagnosis model; the learning resource push module 203 is used to align the features of the learner's knowledge status with the knowledge point labels of the learning interaction data and match them with each other, extract the learning resource text content according to the matching results and push them.

[0044] It is worth mentioning that the method implemented by the interactive language model-based learning resource push device 20 disclosed in the second embodiment of the present invention is as described in the first embodiment above, and therefore will not be described in detail here. Optionally, the various modules and other operations or functions described above in the second embodiment are respectively for implementing the interactive language model-based learning resource push method described in the first embodiment, and the beneficial effects of this embodiment are the same as those of the first embodiment above. For the sake of brevity, they are not repeated here.

[0045] The third embodiment of the present invention also proposes an electronic device 30, for example, including: at least one processing unit 31, and at least one storage unit 32, wherein the storage unit 32 stores a computer program, and when the computer program is executed by the processing unit 31, the processing unit 31 executes the method as described in the first embodiment, and the beneficial effects of the electronic device 30 provided by this embodiment are the same as the beneficial effects of the learning resource push method based on the interactive language model provided by the first embodiment.

[0046] The fourth embodiment of the present invention also provides a computer-readable storage medium 40, on which a computer program is stored. When the program is executed by a processor, the steps of the above method are implemented. The beneficial effects of the computer-readable storage medium 40 provided in this embodiment are the same as the beneficial effects of the learning resource push method based on the interactive language model provided in the first embodiment.

[0047] Among them, computer-readable storage media may include, but are not limited to, any type of disk, including floppy disks, optical disks, DVDs, CD-ROMs, microdrives and magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.

[0048] It should be noted that for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by this application.

[0049] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0050] In the several embodiments provided in this application, it should be understood that the disclosed devices can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some service interface, and the indirect coupling or communication connection of the device or unit can be electrical or other forms.

[0051] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0052] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0053] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a memory and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned memory includes: U disk, read-only memory (ROM), random access memory (RAM), mobile hard disk, magnetic disk, or optical disk, etc., various media that can store program code.

[0054] Those skilled in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be completed by instructing related hardware through a program. The program may be stored in a computer-readable memory, which may include a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0055] The above is only an exemplary embodiment of the present disclosure and cannot be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. After considering the specification and practicing the disclosure herein, those skilled in the art will easily think of the implementation scheme of the present disclosure. This application is intended to cover any variation, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary technical means in the art that are not recorded in the present disclosure. The description and examples are to be regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.

[0056] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0057] It will be easily understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A learning resource push method based on an interactive language model, characterized in that: include: Acquire learners' learning interaction data, perform semantic analysis on the learning interaction data based on a large language model, and extract key features; Performing feature enhancement on the answer text information and knowledge point text information of the key features, constructing a knowledge status diagnosis model based on the enhanced features, and predicting the learner's knowledge status through the forgetting decay coefficient matrix of the knowledge status diagnosis model; The learner's knowledge state and the knowledge point labels of the learning interaction data are feature-aligned and matched with each other, and the learning resource text content is extracted and pushed according to the matching results.

2. The learning resource push method based on the interactive language model according to claim 1 is characterized in that: The acquiring of the learner's learning interaction data includes: Obtain learners' interactive answer records based on the online learning platform database , where the learner's answer record set Includes detailed information about each answer, including question information, knowledge point information, answer time, and whether the answer is correct. Extracting learning resources available on the platform, including teaching videos, course materials, and exercises related to course knowledge points, extracting text content from the learning resources, and saving them as corresponding text files; The answer records of each learner and the learning resources collected by the platform are divided into different files for storage according to teaching video resources, course materials and exercises related to course knowledge points, wherein each file contains the corresponding text file.

3. The learning resource push method based on the interactive language model according to claim 1, characterized in that: After acquiring the learner's learning interaction data, the method further includes: Deleting and completing the answer sequence in the learning interaction data, and sorting them by timestamp; The learning interaction data is encoded using UTF-8 and standardized into text data, and the text data is segmented based on the tokenizer for feature embedding generation by the LLM module; The learning interaction data is divided into several equal parts, part of which is used for model training, part of which is used for model verification, and part of which is used for model testing.

4. The method for pushing learning resources based on an interactive language model according to claim 1, characterized in that: The knowledge state diagnosis model includes a first sub-network model and a second sub-network model. The first sub-network model is used to construct a complete embedding expression of the learner's answer information by combining the text features enhanced by the large language model features, and output a latent feature vector; The second sub-network model is used to perform a decoding operation on the acquired embedded expression and obtain a final prediction result in combination with the forgetting decay coefficient matrix.

5. The method for pushing learning resources based on an interactive language model according to claim 4, characterized in that: The execution steps of the first sub-network model include: Using the pre-trained large language model to extract features from the learner's answer text and knowledge point text information; Obtaining a complete embedding expression based on the feature extraction, and combining the key features extracted by the large language model with other relevant features; A positional encoding is constructed based on the Transformer's self-attention mechanism, and the complete embedded expression is fed into the encoder to capture the complex logical relationship between knowledge points and exercises, thereby outputting a potential feature vector.

6. The method for pushing learning resources based on an interactive language model according to claim 4, characterized in that: The execution steps of the second sub-network model include: Performing a decoding operation based on the latent feature vector, wherein the latent feature of the knowledge point is used as a query function to obtain the learner's knowledge mastery level at different knowledge points; Constructing a forgetting decay coefficient matrix based on the temporal distance between questions to adjust the importance of each time step in the output sequence, and multiplying the output sequence by the forgetting decay coefficient matrix to adjust the output sequence; Based on the predicted probability of the output sequence output knowledge, the potential feature vector is input into the fully connected layer for dimensional feature change, and is converted into a target probability value in combination with the activation function.

7. The method for pushing learning resources based on an interactive language model according to claim 1, characterized in that: The learning resource pushing step includes: Use cosine similarity and other measurement methods to align the learner's mastery of each knowledge point with the knowledge point labels of the learning resources. Based on the alignment results, identify the learning resources that best match the learner's current knowledge status. The weights of all resources are normalized to ensure that the total weight is 1. Based on the normalized weights and matching results, the learning resources with the highest weight are selected for push.

8. A learning resource push device based on an interactive language model, characterized in that: include: A key feature extraction module is used to obtain learners' learning interaction data, perform semantic analysis on the learning interaction data based on a large language model, and extract key features; A knowledge status prediction module is used to enhance the features of the answer text information and knowledge point text information of the key features, build a knowledge status diagnosis model based on the enhanced features, and predict the learner's knowledge status through the forgetting decay coefficient matrix of the knowledge status diagnosis model; The learning resource push module is used to align and match the learner's knowledge status with the knowledge point tags of the learning interaction data, extract the learning resource text content according to the matching results, and push it.

9. An electronic device, characterized in that: The method comprises at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program, and when the computer program is executed by the processing unit, the processing unit performs the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that It stores a computer program executable by an access authentication device. When the computer program runs on the access authentication device, the access authentication device executes the steps of the method according to any one of claims 1 to 7.

Images