A method for solving the cold start problem in using open teaching resources based on semantic rules

Abstract

The present invention relates to the field of data analysis technology, and specifically to a method for solving the cold start problem in the use of open teaching resources based on semantic rules, including: when a new learner or new resource is detected, a lightweight learner-micro OER profile is generated, an initial recommendation list is generated using a heuristic algorithm, and new resources are inserted into the learning path; wherein the lightweight learner-micro OER profile includes: information about the learner. This solution can solve the cold start problem of new learners and new resources in MLaaS, and provide personalized learning paths and resource recommendations to improve the utilization rate and promotion of MLaaS.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to a method for solving a cold start problem in using open teaching resources based on semantic rules. Background Art

[0002] With the development of Internet technology, distance education has gradually become an important teaching method; many education providers have begun to try to deliver knowledge and learning materials remotely through multiple channels; micro-open learning is an emerging trend that allows learners to learn in fragmented time, but faces the problem of data sparsity, that is, the lack of sufficient learner information and the continuous release of new OERs (open educational resources).

[0003] In order to promote micro-open learning, researchers proposed MLaaS (Micro-Learning as a Service), which is a cloud-based system designed to support mobile micro-open learning; MLaaS operates by searching and calling all available OERs and their associated web services. Its core is an adaptive engine that takes into account learners' personal information, immediate time availability, mobile environment characteristics, and the characteristics of micro-OERs to achieve adaptive delivery of learning resources.

[0004] MLaaS aims to provide personalized micro-OERs based on learners' real-time availability and the characteristics of micro-learning scenarios, but the cold-start problem limits the use and promotion of MLaaS, where the cold-start problem refers to the problem that when new users or new projects join, it is difficult to accurately generate personalized recommendations for these new users or new projects (i.e., new learners and new resources) due to the lack of sufficient historical data. In the context of MLaaS, the cold-start problem is particularly prominent because learners in micro-open learning environments are often self-regulated and task-oriented, and new learners are constantly joining or existing learners may frequently switch courses. Summary of the invention

[0005] The present invention aims to provide a method for solving the cold start problem in the use of open teaching resources based on semantic rules, which can solve the cold start problem of new learners and new resources in MLaaS, provide personalized learning paths and resource recommendations, so as to improve the utilization rate and promotion of MLaaS.

[0006] The present invention provides the following basic solution: a method for solving the cold start problem in using open teaching resources based on semantic rules, including the following contents:

[0007] When a new learner or new resource is detected, a lightweight learner-micro OER profile is generated, an initial recommendation list is generated using a heuristic algorithm, and new resources are inserted into the learning path; wherein the lightweight learner-micro OER profile includes: information about the learner.

[0008] Further, the lightweight learner-micro OER profile is divided into: a static part and a dynamic part;

[0009] The static part is a vector that represents the learner's basic personal information and educational information;

[0010] The dynamic part is updated in real time according to the learner's behavior changes and is expressed in the form of a pair, ML j ={MR u ,L j}, where L j ∈L; element MR u Represents the micro-OER labeled by user u, the three-dimensional element L j {P u,j ,TA j ,D j} is the jth learner in the whole micro-learning process, P u,j represents the preference value of learner j for micro-OER u; TA j represents the length of time available to learner j; D j Represents the educational background or domain knowledge of learner j.

[0011] Furthermore, when a new learner or a new resource is detected, a lightweight learner-micro OER profile is generated, an initial recommendation list is generated using a heuristic algorithm, and new resources are inserted into the learning path, including:

[0012] Initialization step: When a new learner or new resource is detected, online computation is started;

[0013] Predict features step: predict learner features;

[0014] Generate candidate paths step: randomly generate several candidate learning paths;

[0015] Evaluation and optimization step: Evaluate each candidate learning path, calculate the degree of violation of the rules, and optimize the candidate learning path through the optimization model;

[0016] Insert new resource step: For the new micro-OER, calculate its similarity with the existing resources, and insert the new OER into the appropriate position according to the preset rules;

[0017] Generate recommendation steps: Generate and optimize the recommendation list and select the optimal learning path.

[0018] Furthermore, the feature prediction step includes:

[0019] Data collection: Obtain information about new learners and their initial interaction data in the system;

[0020] Feature prediction: Use heuristic algorithms to predict the features of new learners based on the information obtained about them;

[0021] Confidence assessment: Assign a confidence level to each predicted feature;

[0022] Dynamic Update: As new learners increase their activity in the MLaaS system, the MLaaS system continuously updates the dynamic part of the lightweight learner-micro OER profile.

[0023] Furthermore, the step of generating candidate paths includes: randomly generating a number of candidate learning paths using existing learning resources according to the predicted characteristics of the new learner and its lightweight learner-micro OER profile.

[0024] Furthermore, the relationship between learning resources in the MLaaS system includes:

[0025] A constituent OER is an inclusion relationship between a non-micro-OER and one or more micro-OERs;

[0026] The requirement sequence indicates a strong sequential relationship between two items, i.e. the former must be learned before the latter; the items include: OER and / or micro-OER;

[0027] The recommendation sequence represents a weak sequential relationship between two items, suggesting that the previous item should be learned before the latter item.

[0028] Further, the evaluation and optimization step includes: setting heuristic rules according to the relationship between learning resources in the MLaaS system, including:

[0029] For each pair of micro-learning resources u and v, the specifics include:

[0030] (1) If there is a demand sequence (MR) between two micro-learning resources u and v u ,MR v ), then the resource u that needs to be learned first has a higher priority, that is, the precondition MR u Place on top;

[0031] (2) If learner j prefers u to v, that is, P u,j >P v,j , then u has a higher priority, that is, MR u Put it in MR v above;

[0032] (3) If in absolute value, learner j’s confidence in u is CD(P u ,j) is greater than the confidence CD(P v,j), then u has a higher priority;

[0033] (4) If there is a recommendation sequence (MR) between u and v u ,MR v ), then u has a higher priority when recommending micro-learning resources;

[0034] (5) Micro-learning resources that are more relevant to the learners’ educational background are given higher priority;

[0035] (6) If there are obvious disciplinary differences between u and v, micro-learning resources with stronger disciplinary relevance will be given priority;

[0036] (7) If none of rules (1)-(6) apply, the micro-learning resources are randomly arranged.

[0037] Furthermore, the evaluation and optimization step further includes: using an optimization model to evaluate each candidate learning path, calculating the degree of violation of the rule, and optimizing the candidate learning path;

[0038] The input of the optimization model includes: an array of learners' preference values ​​for resources, the learners' current available time, external environment impact value, preference value, confidence of time and environment value, difficulty of each resource, and heuristic rules 1-7;

[0039] The optimization model evaluates each candidate learning path based on the input, calculates the degree of violation of the rules, optimizes the learning path, and outputs the optimal learning path as the sorting sequence of micro-learning resources;

[0040] The data processing process of the optimization model includes:

[0041] Step 1: Get several randomly generated candidate learning paths, take each learning path as a chromosome, and several candidate learning paths form a chromosome set;

[0042] Step 2: Initialize the number of iterations n = 1;

[0043] Step 3: Calculate the initial fitness η of each chromosome;

[0044] Specifically, the fitness η is calculated for each micro-learning resource u in each chromosome k in the chromosome set:

[0045]

[0046] Among them, α, β, γ, and δ are weights, and α>β>γ>δ;

[0047] in

[0048]

[0049] Step 4: Determine whether the number of iterations n is greater than the preset number of iterations. If so, execute Step 5. If not, generate a new chromosome and update n=n+1, and then execute Step 4;

[0050] The generated new chromosomes include: based on the chromosome set, generating a new chromosome k', for each new chromosome, judging whether the time length of the first micro-OER of chromosome k' is less than or equal to the learner's current available time Ta j , if yes, then keep chromosome k', if no, then do not keep chromosome k';

[0051] Calculate the fitness η' of the new chromosome k' and determine whether η' is less than η. If so, replace chromosome k with the new chromosome k'; if not, retain chromosome k;

[0052] Step 5: Select M optimal chromosomes in the chromosome set. Specifically, sort the chromosomes in the chromosome set from small to large according to fitness, and select the first M chromosomes as the optimal chromosomes;

[0053] Step 6: Select the chromosome that meets the learner's current available time from the optimal chromosome as the recommended learning path;

[0054] Step 7: Output the recommended learning path as a sorted sequence of micro-learning resources, and take the first micro-learning resource in the recommended learning path as the initial recommendation.

[0055] Furthermore, the step of inserting new resources includes: when a new micro-OER is released, using the insertion model to calculate the similarity between the new micro-OER and the existing micro-OERs; according to the heuristic rules, the MLaaS system inserts the new micro-OERs at the appropriate position;

[0056] Insert the model input, including: New Micro OER MR n , existing micro-OERs, established learning paths, and heuristic rules in the MLaaS system;

[0057] The insertion model calculates the similarity between the new micro-OER and the existing micro-OERs based on the input, inserts the new OERs into the learning path according to the heuristic rules, and outputs the learning path containing the new micro-OER, in which the new micro-OER is inserted at the appropriate position.

[0058] Furthermore, the data processing process of the insertion model includes:

[0059] Step 1: Calculate the new micro-OER MR n Similarity to existing micro-OERs:

[0060]

[0061] Where t1 is the current time and t2 is the MR u Release time; j is extracted from the real-time MLaaS system usage j The constant c keeps the denominator not equal to 0;

[0062] Step 2: Use the K nearest neighbor algorithm to map the new micro OER MR n Clustered into micro-OERs groups whose similarity meets the preset similarity threshold, the corresponding neighbors form a set G; let MR g ∈G, thus obtaining the prediction equation of the new micro-OER preference value:

[0063]

[0064] Step 3: Check semantic relationships:

[0065] Investigating new micro-OER MR n The semantic relationship between its neighbors in set G;

[0066] Retrieve the established learning paths and select the path containing the neighbor micro-OER, or use the optimization model to generate new candidate learning paths;

[0067] Step 4: Process each neighbor:

[0068] For each neighbor micro-OER MR g , select Include MR g The learning path of segmentation includes MR g and several micro-OER paragraphs before and after it;

[0069] Check the new micro OER MR n The semantic relationship between the micro-OERs in the segmented paragraphs;

[0070] Step 5: Insert the new OER MR n :

[0071] If there is a demand sequence, the new micro OER MR n inserted between two micro-OERs;

[0072] If there is a recommended sequence, the new micro OER MR n Put it in the right place;

[0073] If there is no relationship, the new micro OER MR n Placed in parallel with existing micro-OERs;

[0074] Step 6: Use the prediction equation of preference value to evaluate the insertion MR n The predicted preference values ​​of all micro-OERs are shown below;

[0075] Step 7: Apply heuristic rule 2 to re-evaluate the new learning path;

[0076] Step 8: Use the hill climbing algorithm to calculate the fitness of the new learning path;

[0077] Step 9: If the fitness of the new learning path is greater than the fitness of the existing learning path, the new learning path is used to replace the existing learning path;

[0078] Step 10: Output the new learning path, which contains the newly inserted MR n learning path.

[0079] Beneficial effects: This scheme uses online computing to overcome the cold start problem in micro-open learning; specifically, by combining lightweight learner-micro-OER profiles and heuristic algorithms, the MLaaS system is able to provide high-quality personalized recommendations for new users (new learners) when data is scarce; a lightweight learner-micro-OER profile is designed to quickly generate preliminary recommendations when data is scarce, and by randomly generating candidate learning paths and continuously optimizing them, the optimal learning path is finally selected as the initial recommendation for new users.

[0080] This solution can solve the cold start problem of new learners and new resources in MLaaS, and provide personalized learning paths and resource recommendations to improve the usage and promotion of MLaaS. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 It is a schematic diagram of an MLaaS system in an embodiment of a method for solving a cold start problem in using open teaching resources based on semantic rules of the present invention;

[0082] Figure 2 A schematic diagram of the Precision values ​​of different recommendation systems on the cold start problem of new users in an embodiment of the method for solving the cold start problem in using open teaching resources based on semantic rules of the present invention;

[0083] Figure 3 A schematic diagram of Recall values ​​of different recommendation systems on the cold start problem of new users in an embodiment of the method for solving the cold start problem in using open teaching resources based on semantic rules of the present invention;

[0084] Figure 4A schematic diagram of average violation degrees of learning paths generated by different methods in an embodiment of the method for solving the cold start problem in using open teaching resources based on semantic rules of the present invention;

[0085] Figure 5 A schematic diagram of average violation degrees of optimal learning paths generated by different methods in an embodiment of the method for solving the cold start problem in using open teaching resources based on semantic rules of the present invention; DETAILED DESCRIPTION

[0086] The following is further described in detail through specific implementation methods:

[0087] This embodiment provides a method for solving the cold start problem in using open teaching resources based on semantic rules, including the following contents:

[0088] Simplifying the learner model in the MLaaS system and generating a lightweight learner-micro OER configuration file; wherein the lightweight learner-micro OER configuration file includes: information of the learner, specifically, the lightweight learner-micro OER configuration file is divided into: a static part and a dynamic part;

[0089] The static part is a vector that represents the learner's basic personal information and educational information, usually including demographic information such as age, gender, occupation, and educational background. This information is relatively fixed and provides a basic framework to understand the learner's basic situation, which helps to build a preliminary learner profile, so that the MLaaS system can still make basic recommendation decisions based on this basic information when there is a lack of more detailed data. Figure 1 As shown, Figure 1 The names of the modules are as follows:

[0090] Offline Computation:

[0091] Historical Learner Information Gathering: Historical Learner Information Gathering;

[0092] Learner Behavior Tracking: Learner behavior tracking;

[0093] Pre-Knowledge Assessment: Pre-knowledge assessment;

[0094] Course Module Extracting: Course module extraction;

[0095] OER Screening: Open Educational Resources Screening;

[0096] Time Length: time length;

[0097] Micro OER: micro open educational resources;

[0098] No Micro OER: No micro open educational resources;

[0099] Learner Modelling: Learner modeling;

[0100] Educational Data Mining: Educational Data Mining;

[0101] Resource Segmentation: Resource segmentation;

[0102] Meta Data Standard: Metadata standard;

[0103] Micro OER Representation: Micro open educational resources representation;

[0104] Adaptive Engine:

[0105] Educational Data Mining and Learning Analytics: Educational Data Mining and Learning Analytics;

[0106] Micro OER Clustering: Micro open educational resources clustering;

[0107] Resource Correlation Measuring: Resource Correlation Measuring;

[0108] Learning Path Identification: Learning Path Identification;

[0109] Adaptive Micro OER Delivery: Adaptive Micro OER Delivery;

[0110] Real-time Learner Data Retrieval: Real-time learner data retrieval;

[0111] Real-Time Profile Updating: Real-Time Profile Updating;

[0112] Recall and Re-evaluation Service: Recall and Re-evaluation Service;

[0113] New Micro OER Measuring: New Micro OER Measuring;

[0114] Collaborative Filtering Service: Collaborative filtering service;

[0115] Place finding in Established Learning Path: Place finding in Established Learning Path;

[0116] Lightweight Learner-micro OER Profile:Lightweight Learner-micro OER Profile;

[0117] Learner Feature Prediction: Learner feature prediction;

[0118] Cold Start Engine: Cold Start Engine;

[0119] Online Computation:

[0120] User Interface: User interface.

[0121] The dynamic part is updated in real time according to the learner's behavior changes and is expressed in the form of a pair, namely ML j ={MR u ,L j}, where L j ∈L; element MR u represents the micro-OER tagged by user u, which is a specific instance of the enhanced micro-OER ontology; while the three-dimensional element L j {P u,j ,TA j ,D j} is exclusive to the jth learner in the entire micro-learning process, P u,j represents the preference value of learner j for micro-OER u; TA j represents the length of time available to learner j; D j represents the educational background or domain knowledge of learner j; confidence (CD) represents the subjective relevance profile of each element in the lightweight profile; for example, if the learner directly provides explicit information, displayed as a numerical value rather than a fuzzy range, its relevant confidence is set to full; if any information change of the jth learner during the micro-open learning process is detected, the MLaaS system will automatically set the L in the lightweight learner-micro-OER profile to full. j Part. The variables in the lightweight learner-micro OER profile are shown in Table 1 below;

[0122] Table 1: Symbol table

[0123]

[0124]

[0125] Micro-OER (micro open educational resources) is a subset of OERs (open educational resources) and refers to learning units or activities that can be completed in a short period of time (usually less than 15 minutes).

[0126] When a new learner or new resource is detected, a lightweight learner-micro-OER profile is generated, a heuristic algorithm is used, offline calculations are performed, an initial recommendation list is generated, and new resources, i.e., newly released micro-OERs, are inserted into the learning path;

[0127] Specifically,

[0128] Initialization step: When a new learner or new resource is detected, online computation is started;

[0129] Predict feature step: predict learner features in order to activate the sparse data filling process; the sparse data filling process aims to build a preliminary learner model for new users using limited data information; this helps to provide effective recommendations in the absence of a large amount of historical behavior data;

[0130] In order to generate personalized recommendations quickly, even for new users or when data is scarce, it is necessary to predict the basic characteristics and preferences of learners. Heuristic algorithms and rule-based methods are used to predict learners' attributes, including but not limited to age, gender, educational background, areas of interest, etc.

[0131] Specifically, data collection: obtaining information about new learners (i.e., new users), such as basic information provided during registration (static part), and initial interaction data of new learners in the system (dynamic part);

[0132] Feature prediction: Use heuristic algorithms to predict the features of new learners based on the information obtained about them; for example, if a user clicks on a certain micro-OER, the system may infer the user’s interest in that field;

[0133] Confidence evaluation: For each predicted feature, a confidence score (CD) is assigned; if the new learner directly provides explicit information (such as a rating), its confidence score will be set to a high value;

[0134] Dynamic Update: As new learners increase their activity in the MLaaS system, the MLaaS system will continuously update the dynamic part of the lightweight learner-micro OER profile to reflect the latest user behavior and preferences; based on the predicted features and updated profile, the system will generate a list of recommendations. These recommendations will be adjusted according to the user's real-time behavior to improve the accuracy and relevance of the recommendations.

[0135] Generating candidate paths: randomly generating several candidate learning paths; based on the predicted characteristics of new learners and their lightweight learner-micro OER profiles, using existing learning resources, randomly generating several candidate learning paths;

[0136] Evaluation and optimization step: Evaluate each candidate learning path, calculate the degree of violation of the rules, and optimize the candidate learning path through an optimization model, such as GA or EP;

[0137] Specifically, when dealing with large amounts of OER, the MLaaS system focuses on the following three main relationships:

[0138] A constituent OER (ConsistsOf) is a containment relationship between a non-micro-OER and one or more micro-OERs; for example, the original OER Ra can be split into several micro-OERs, one of whose descendants is MRu, constituting (Ra, MRu); in addition, a micro-OER can also be further subdivided to form smaller micro-OERs;

[0139] RequiredSequence represents a strong sequential relationship between two items (OER and / or micro-OER), i.e. the former item must be studied before the latter item, based on a predefined course outline;

[0140] Recommended Sequence represents a weak sequential relationship between two items. It is not mandatory and is usually based on teacher suggestions or course settings. It is recommended that the previous item be learned before the next item.

[0141] Required sequence and recommended sequence, both of which can be inherited by descendants of entities in the extended micro-OER ontology; for example, in the recommended sequence (R a , R b ) is recommended in R b Learn R before a , for MR u ∈R a and MR v ∈R b There is also a recommended sequence (MR u , MR v); In addition, this inheritance is effective considering that there is a certain relationship between two micro-OERs; Once the MLaaS system obtains the dynamic part of each lightweight learner-micro-OER profile, it will quickly check the final predicted preference value and confidence of the micro-OER node; The MLaaS system will reject nodes whose confidence does not meet the minimum acceptance criterion; In order to determine the MR u and MR v The position in the recommendation list, so the following heuristic rules are set based on the above three types of relationships:

[0142] For each pair of micro-learning resources u and v, the specifics include:

[0143] 1. If there is a demand sequence (MR) between two micro-learning resources u and v u ,MR v ), then the resource u that needs to be learned first has a higher priority, that is, the precondition MR u Place on top;

[0144] 2. If learner j prefers u to v, that is, P u,j >P v,j , then u has a higher priority, that is, MR u Put it in MR v above;

[0145] 3. If in absolute value, learner j's confidence in u is CD(P u ,j) is greater than the confidence CD(P v ,j), then u has a higher priority;

[0146] 4. If there is a recommendation sequence (MR) between u and v u ,MR v ), then u has a higher priority when recommending micro-learning resources;

[0147] 5. Micro-learning resources that are more relevant to the learner’s educational background have higher priority;

[0148] 6. If there are obvious disciplinary differences between u and v, micro-learning resources with stronger disciplinary relevance will be given priority;

[0149] 7. If none of rules 1-6 apply, the micro-learning resources are randomly arranged.

[0150] The input of the optimization model includes: learner preference value array for resources (learnerPreferences), learner current available time (availableTime), external environment influence value (environmentInfluence), preference value, confidence of time and environment value (credibility), difficulty of each resource (resourceDifficulty), heuristic rules 1-7;

[0151] The optimization model evaluates each candidate learning path based on the input, calculates the degree of violation of the rules, optimizes the learning path, and outputs the optimal learning path, that is, the output: the sorted sequence of micro-learning resources;

[0152] The data processing process of the optimization model includes:

[0153] Step 1: Get several randomly generated candidate learning paths, take each learning path as a chromosome, and several candidate learning paths form a chromosome set;

[0154] Step 2: Initialize the number of iterations n = 1;

[0155] Step 3: Calculate the initial fitness η of each chromosome;

[0156] Specifically, the fitness η is calculated for each micro-learning resource u in each chromosome k in the chromosome set:

[0157]

[0158] Among them, α, β, γ, and δ are weights, and α>β>γ>δ;

[0159] in

[0160]

[0161] Step 4: Determine whether the number of iterations n is greater than the preset number of iterations. If so, execute Step 5. If not, generate a new chromosome and update n=n+1, and then execute Step 4;

[0162] The generated new chromosomes include: based on the chromosome set, generating a new chromosome k', for each new chromosome, judging whether the time length of the first micro-OER of chromosome k' is less than or equal to the learner's current available time Ta j , if yes, then keep chromosome k', if no, then do not keep chromosome k';

[0163] Calculate the fitness η' of the new chromosome k' and determine whether η' is less than η. If so, replace chromosome k with the new chromosome k'; if not, retain chromosome k;

[0164] Step 5: Select M optimal chromosomes in the chromosome set. Specifically, sort the chromosomes in the chromosome set from small to large according to fitness, and select the first M chromosomes as the optimal chromosomes;

[0165] Step 6: Select the chromosome that meets the learner's current available time from the optimal chromosome as the recommended learning path;

[0166] Step 7: Output the recommended learning path as the sorted sequence of micro-learning resources, and take the first micro-learning resource in the recommended learning path as the initial recommendation. By randomly generating candidate learning paths and continuously optimizing them, the optimal learning path is finally selected as the initial recommendation for new users.

[0167] Insert new resource step: For the new micro-OER, calculate its similarity with the existing resources, and insert the new OER into the appropriate position according to the preset rules;

[0168] Specifically, when a new micro-OER is released, the MLaaS system will use the insertion model to calculate the similarity between the new micro-OER and the existing micro-OERs, which helps to determine the appropriate position of the new resource in the existing learning path; according to the heuristic rules, the MLaaS system will insert the new micro-OERs at the appropriate position; for example, if there is a recommended sequence, the new micro-OER will be inserted into the corresponding position;

[0169] Insert the model input, including: New Micro OER (MR n ), existing micro-OERs, established learning paths, and heuristic rules in the MLaaS system;

[0170] The insertion model calculates the similarity between the new micro-OER and the existing micro-OERs based on the input, inserts the new OERs into the learning path according to the heuristic rules, and outputs the learning path containing the new micro-OER, in which the new micro-OER is inserted at the appropriate position;

[0171] The data processing for inserting into the model includes:

[0172] Step 1: Calculate the new micro-OER MR n Similarity to existing micro-OERs:

[0173]

[0174] Where t1 is the current time and t2 is the MR u Release time; jis extracted from the real-time MLaaS system usage j The operation time of the constant c keeps the denominator not equal to 0; the above calculation method uses the time decay factor and penalty term to improve the similarity calculation based on Euclidean distance. The former calculates the accumulation and decay of learners' interests, while the latter breaks the filter bubble.

[0175] Step 2: Use the K nearest neighbor algorithm to map the new micro OER MR n Clustered into micro-OERs groups whose similarity meets the preset similarity threshold, the corresponding neighbors form a set G; let MR g ∈G, thus obtaining the prediction equation of the new micro-OER preference value:

[0176]

[0177] Step 3: Check semantic relationships:

[0178] Investigating new micro-OER MR n The semantic relationship between its neighbors in set G;

[0179] Retrieve the established learning paths and select the path containing the neighbor micro-OER, or use the optimization model to generate new candidate learning paths;

[0180] Step 4: Process each neighbor:

[0181] For each neighbor micro-OER MR g , select Include MR g The learning path of segmentation includes MR g And several micro-OER paragraphs before and after it.

[0182] Check the new micro OER MR n The semantic relationship between the micro-OERs in the segmented paragraphs;

[0183] Step 5: Insert the new OER MR n :

[0184] If there is a demand sequence, the new micro OER MR n inserted between two micro-OERs;

[0185] If there is a recommended sequence, the new micro OER MR n Put it in the right place;

[0186] If there is no relationship, the new micro OER MR n Placed in parallel with existing micro-OERs;

[0187] Step 6: Use the prediction equation of preference value to evaluate the insertion MRn The predicted preference values ​​of all micro-OERs are shown below;

[0188] Step 7: Apply heuristic rule 2 to re-evaluate the new learning path;

[0189] Step 8: Use the hill climbing algorithm to calculate the fitness of the new learning path;

[0190] Step 9: If the fitness of the new learning path is greater than the fitness of the existing learning path, the new learning path is used to replace the existing learning path;

[0191] Step 10: Output the new learning path, which contains the newly inserted MR n learning path.

[0192] By calculating the similarity between the new micro-OER and the existing micro-OERs, it is inserted into the appropriate position to optimize the existing learning path.

[0193] Generate recommendation steps: Generate and optimize the recommendation list and select the optimal learning path.

[0194] This scheme exploits online computing to overcome the cold-start problem in micro-open learning; specifically, by combining lightweight learner-micro-OER profiles and heuristic algorithms, the MLaaS system is able to provide high-quality personalized recommendations for new users (new learners) in the presence of scarce data.

[0195] Experimental results show that this solution outperforms existing recommendation systems in terms of recommendation performance and can effectively handle the cold start problem of new users and new resources:

[0196] Performance comparison: Figure 2 As shown in the figure, through experimental evaluation, it is proved that the MLaaS method is superior to other recommendation systems in terms of indicators such as Precision and Recall, especially when dealing with the cold start problem of new users;

[0197] Specifically, the experiments set up different recommender systems, including context-aware recommender system (N-based), ACOS, genetic algorithm (GA), and evolutionary programming (EP);

[0198] Recommendation list size: N varies from 1 to 20 to observe the impact of different recommendation list sizes on precision;

[0199] Evaluation Metrics: Use precision as the evaluation metric, which represents the proportion of relevant items in the recommendation list;

[0200] Results, such as Figure 2 As shown, Figure 2The x-axis is N, which indicates the size of the recommendation list; the y-axis is the precision value, which indicates the proportion of relevant items in the recommendation list; the curve is the trend of the precision value of different recommendation systems with N: context-aware recommendation system (N-based), when no context information is provided (i.e., cold start situation), this recommendation system cannot generate competitive and accurate results; ACOS, when N is greater than 18, the quality of recommendations produced by ACOS and GA is very close, but the initial recommendation quality of ACOS is relatively poor; NHSM, the use of NHSM can significantly improve the performance of user-based and item-based recommendation systems, but in terms of precision, it cannot compete with GA-based and EP-based MLaaS methods; GA and EP (this scheme), regardless of the size of N (N≤20), GA-based and EP-based MLaaS methods perform best in terms of precision of the overall results; when N=15, these two methods perform well in retrieving relevant micro-OERs. Among all the recommendation systems, this scheme performs best in dealing with the cold start problem of new users, especially when N=15, it can retrieve relevant micro-OERs more effectively.

[0201] Significance test: Figure 3 As shown, the paired t-test statistical method is used to verify that the performance difference between this scheme and the existing method is statistically significant;

[0202] Specifically, the experiment sets up different recommender systems and compares various recommender systems, including context-aware recommender system (N-based), ACOS, genetic algorithm (GA), and evolutionary programming (EP);

[0203] Recommendation list size: N varies from 1 to 20 to observe the impact of different recommendation list sizes on Recall;

[0204] Evaluation indicators: Recall is used as the evaluation indicator, which indicates the proportion of recommended items among all related items;

[0205] Results, such as Figure 3As shown in the figure, x-axis: N, indicating the size of the recommendation list; y-axis: Recall value, indicating the proportion of recommended items in all related items; curve: Recall value of different recommendation systems changing with N; NHSM, when N is less than 5, the Recall value generated by NHSM is the highest; as N increases, this scheme gradually takes the lead; ACOS is the only recommendation system in which Precision and Recall improve simultaneously with the increase of N, however, the Recall growth curve of ACOS is always lower than that of NHSM; this scheme performs best in Recall when N increases; compared with NHSM, this scheme shows higher Recall values ​​at larger N values; other recommendation systems, user clustering recommendation system (U-Clustering) and other typical similarity calculation models (such as NHSM and ACOS) perform worse than this scheme in terms of Recall.

[0206] Among all the recommendation systems, our schemes perform best when dealing with the cold start problem of new users, especially at larger N values, where they are able to retrieve relevant micro-OERs more effectively. At smaller N values, NHSM generates higher Recall values, but as N increases, its performance is gradually surpassed by GA and EP. Although the Recall value of ACOS increases with the increase of N, its growth curve is always lower than that of NHSM and GA / EP methods. Our scheme not only performs well in Precision, but also has a significant advantage in Recall, thus providing more comprehensive and effective recommendations.

[0207] Learning path quality: Figure 4 As shown in the figure, the quality of the learning path was evaluated by inserting the newly released micro-OERs and generating new learning paths. The results showed that the learning path generated by this scheme has a lower violation degree (VD), which is better than other methods.

[0208] Specifically, the experimental settings,different fields, involve micro-OERs in three fields: information technology, business, and social sciences;,the present scheme, the shortest path (SP), and the,competency-based approach (C-based) are adopted;,wherein SP is to find the shortest path from one point to another, but does not consider,inserting the newly released micro-OER into the existing learning path;,C-based is to reconstruct the learning path based on the,learner's skill level, and does not directly deal with the,insertion problem of the newly released micro-OER;

[0209] Evaluation metric: Violation Degree (VD), which is used to measure whether the relationship between consecutive micro-OER pairs in the learning path violates the pre-set rules. A lower VD value means a learning path that is more compliant with the rules.

[0210] Results, such as Figure 4As shown, the x-axis: different fields, including information technology, business and social sciences; the y-axis: average violation degree (VD), which indicates the degree of violation of preset rules in the learning path; the bar chart: shows the average VD value of all learning paths generated by each method in each field; Figure 4 The average violation degree (VD) of the learning paths generated by different methods is shown. The learning paths are generated by inserting newly released micro open educational resources (micro OERs) in three fields: information technology, business, and social sciences. Specifically, the results of the experiments and characterizations are as follows: The proposed scheme performs best in all subject areas, and the average VD value of the generated learning paths is the lowest, which shows that the proposed scheme can place the newly released micro OERs more appropriately and better follow the relevant principles. SP method: Although it tries to find a reasonable path, it is difficult for the SP method to identify a suitable learning path due to the large number of recommended sequences, resulting in a high average VD value. Capability-based method (C-based): Although it also tries to create a new learning path for each tested micro OER, it is not as efficient as the proposed scheme due to the long calculation time, and the average VD value of the generated learning paths is also relatively high. Domain differences: In the field of information technology, since this field contains more systematic knowledge areas, the number of micro OERs involved in the learning paths is large, which leads to a large variation in the VD value; while in the business field, due to the more complex associations between learning resources, all methods produce high VD values. Compared with the other two methods, this scheme can generate learning paths with lower violation levels in all subject areas, showing its advantage in dealing with newly released micro-OERs. In different subject areas, the characteristics and complexity of learning resources are different, which leads to differences in the difficulty of building learning paths. For example, in the field of information technology, a more systematic knowledge system needs to be covered, which leads to a larger number of micro-OERs in the learning path, thereby increasing the complexity of path construction. In the business field, the VD value is further improved because there are more complex correlations between learning resources.

[0211] Figure 5The figure shows the average violation degree (VD) of the best learning paths generated by different methods. Specifically, the figure compares the VD values ​​of the best learning paths generated by the MLaaS method, the shortest path (SP) method, and the capability-based method (C-based method) for each test micro-OER in the three fields of information technology, business, and social sciences. For each method, a learning path containing a test micro-OER is selected, and the newly generated learning path with the lowest VD value is selected from them. In the field of information technology, the VD value of the learning path generated by this scheme is significantly lower than that of the other two methods, indicating that it can more effectively insert the newly released micro-OER into the learning path while maintaining a low rule violation degree. All methods in the business field produced higher VD values, which may be due to the more complex associations between learning resources in this field, such as some micro-OERs have prerequisites, or multiple micro-OERs are closely related. The social science field also shows a similar result pattern as the information technology field, that is, the MLaaS method outperforms the other two methods. In summary, through Figure 5 It can be observed that the proposed scheme performs well in following relevant principles when dealing with newly released micro-OERs, especially in the fields of information technology and social sciences, where the generated learning paths are more appropriate than the shortest path (SP) approach and the competency-based approach (C-based approach).

[0212] The above is only an embodiment of the present invention. The common sense such as the known specific structure and characteristics in the scheme is not described in detail here. The ordinary technicians in the relevant field know all the common technical knowledge in the technical field of the invention before the application date or priority date, can obtain all the existing technologies in the field, and have the ability to apply the conventional experimental means before that date. The ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can be made, which should also be regarded as the protection scope of the present invention, which will not affect the effect of the implementation of the present invention and the practicality of the patent. The protection scope required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A method for solving the cold start problem in using open teaching resources based on semantic rules, characterized in that: It includes the following: When a new learner or new resource is detected, a lightweight learner-micro OER profile is generated, a heuristic algorithm is used to generate an initial recommendation list, and new resources are inserted into the learning path; The lightweight learner-micro OER profile includes: information about the learner.

2. The method for solving the cold start problem in using open teaching resources based on semantic rules according to claim 1 is characterized in that: The lightweight learner-micro OER profile is divided into: a static part and a dynamic part; The static part is a vector that represents the learner's basic personal information and educational information; The dynamic part is updated in real time according to the learner's behavior changes and is expressed in the form of a pair, ML j ={MR u ,L j }, where L j ∈L; element MR u Represents the micro-OER labeled by user u, the three-dimensional element L j {P u,j ,TA j ,D j } is the jth learner in the whole micro-learning process, P u,j represents the preference value of learner j for micro-OERu; TA j represents the length of time available to learner j; D j Represents the educational background or domain knowledge of learner j.

3. The method for solving the cold start problem in using open teaching resources based on semantic rules according to claim 2 is characterized in that: When a new learner or a new resource is detected, a lightweight learner-micro OER profile is generated, an initial recommendation list is generated using a heuristic algorithm, and new resources are inserted into the learning path, including: Initialization step: When a new learner or new resource is detected, online computation is started; Predict features step: predict learner features; Generate candidate paths step: randomly generate several candidate learning paths; Evaluation and optimization step: Evaluate each candidate learning path, calculate the degree of violation of the rules, and optimize the candidate learning path through the optimization model; Insert new resource step: For the new micro-OER, calculate its similarity with the existing resources, and insert the new OER into the appropriate position according to the preset rules; Generate recommendation steps: Generate and optimize the recommendation list and select the optimal learning path.

4. The method for solving the cold start problem in using open teaching resources based on semantic rules according to claim 3 is characterized in that: The feature prediction step comprises: Data collection: Obtain information about new learners and their initial interaction data in the system; Feature prediction: Use heuristic algorithms to predict the features of new learners based on the information obtained about them; Confidence assessment: Assign a confidence level to each predicted feature; Dynamic Update: As new learners increase their activity in the MLaaS system, the MLaaS system continuously updates the dynamic part of the lightweight learner-micro OER profile.

5. The method for solving the cold start problem in using open teaching resources based on semantic rules according to claim 4 is characterized in that: The step of generating candidate paths includes: randomly generating a number of candidate learning paths using existing learning resources according to the predicted characteristics of the new learner and its lightweight learner-micro OER profile.

6. The method for solving the cold start problem in using open teaching resources based on semantic rules according to claim 5 is characterized in that: The relationship between learning resources in the MLaaS system includes: A constituent OER is an inclusion relationship between a non-micro-OER and one or more micro-OERs; The requirement sequence indicates a strong sequential relationship between two items, i.e. the former must be learned before the latter; the items include: OER and / or micro-OER; The recommendation sequence represents a weak sequential relationship between two items, suggesting that the previous item should be learned before the latter item.

7. The method for solving the cold start problem in using open teaching resources based on semantic rules according to claim 6 is characterized in that: The evaluation and optimization steps include: setting heuristic rules according to the relationship of learning resources in the MLaaS system, including: For each pair of micro-learning resources u and v, the specifics include: (1) If there is a demand sequence (MR) between two micro-learning resources u and v u ,MR v ), then the resource u that needs to be learned first has a higher priority, that is, the precondition MR u Place on top; (2) If learner j prefers u to v, that is, P u,j >P v,j , then u has a higher priority, that is, MR u Put it in MR v above; (3) If in absolute value, learner j’s confidence in u is CD(P u ,j) is greater than the confidence CD(P v ,j), then u has a higher priority; (4) If there is a recommendation sequence (MR) between u and v u ,MR v ), then u has a higher priority when recommending micro-learning resources; (5) Micro-learning resources that are more relevant to the learners’ educational background are given higher priority; (6) If there are obvious disciplinary differences between u and v, micro-learning resources with stronger disciplinary relevance will be given priority; (7) If none of rules (1)-(6) apply, the micro-learning resources are randomly arranged.

8. The method for solving the cold start problem in using open teaching resources based on semantic rules according to claim 7 is characterized in that: The evaluation and optimization step further includes: using an optimization model to evaluate each candidate learning path, calculating the degree of violation of the rule, and optimizing the candidate learning path; The input of the optimization model includes: an array of learners' preference values ​​for resources, the learners' current available time, external environment impact value, preference value, confidence of time and environment value, difficulty of each resource, and heuristic rules 1-7; The optimization model evaluates each candidate learning path based on the input, calculates the degree of violation of the rules, optimizes the learning path, and outputs the optimal learning path as the sorting sequence of micro-learning resources; The data processing process of the optimization model includes: Step 1: Get several randomly generated candidate learning paths, take each learning path as a chromosome, and several candidate learning paths form a chromosome set; Step 2: Initialize the number of iterations n = 1; Step 3: Calculate the initial fitness η of each chromosome; Specifically, the fitness η is calculated for each micro-learning resource u in each chromosome k in the chromosome set: Among them, α, β, γ, and δ are weights, and α>β>γ>δ; in Step 4: Determine whether the number of iterations n is greater than the preset number of iterations. If so, execute Step 5. If not, generate a new chromosome and update n=n+1, and then execute Step 4; The generated new chromosomes include: based on the chromosome set, generating a new chromosome k', for each new chromosome, judging whether the time length of the first micro-OER of chromosome k' is less than or equal to the learner's current available time Ta j , if yes, then keep chromosome k', if no, then do not keep chromosome k'; Calculate the fitness η' of the new chromosome k' and determine whether η' is less than η. If so, replace chromosome k with the new chromosome k'; if not, retain chromosome k; Step 5: Select M optimal chromosomes in the chromosome set. Specifically, sort the chromosomes in the chromosome set from small to large according to fitness, and select the first M chromosomes as the optimal chromosomes; Step 6: Select the chromosome that meets the learner's current available time from the optimal chromosome as the recommended learning path; Step 7: Output the recommended learning path as a sorted sequence of micro-learning resources, and take the first micro-learning resource in the recommended learning path as the initial recommendation.

9. The method for solving the cold start problem in using open teaching resources based on semantic rules according to claim 8, characterized in that: The step of inserting new resources includes: when a new micro-OER is released, using an insertion model to calculate the similarity between the new micro-OER and existing micro-OERs; according to the heuristic rule, the MLaaS system inserts the new micro-OERs at an appropriate position; Insert the model input, including: New Micro OER MR n , existing micro-OERs, established learning paths, and heuristic rules in the MLaaS system; The insertion model calculates the similarity between the new micro-OER and the existing micro-OERs based on the input, inserts the new OERs into the learning path according to the heuristic rules, and outputs the learning path containing the new micro-OER, in which the new micro-OER is inserted at the appropriate position.

10. The method for solving the cold start problem in using open teaching resources based on semantic rules according to claim 9, characterized in that: The data processing process of the insertion model includes: Step 1: Calculate the new micro-OER MR n Similarity to existing micro-OERs: Where t1 is the current time and t2 is the MR u Release time; j is extracted from the real-time MLaaS system usage j The constant c keeps the denominator not equal to 0; Step 2: Use the K nearest neighbor algorithm to map the new micro OER MR n Clustered into micro-OERs groups whose similarity meets the preset similarity threshold, the corresponding neighbors form a set G; let MR g ∈G, thus obtaining the prediction equation of the new micro-OER preference value: Step 3: Check semantic relationships: Investigating new micro-OER MR n The semantic relationship between its neighbors in set G; Retrieve the established learning paths and select the path containing the neighbor micro-OER, or use the optimization model to generate new candidate learning paths; Step 4: Process each neighbor: For each neighbor micro-OER MR g , select Include MR g The learning path of segmentation includes MR g and several micro-OER paragraphs before and after it; Check out the new micro OER MR n The semantic relationship between the micro-OERs in the segmented paragraphs; Step 5: Insert the new OER MR n : If there is a demand sequence, the new micro OER MR n inserted between two micro-OERs; If there is a recommended sequence, the new micro OER MR n Put it in the right place; If there is no relationship, the new micro OER MR n Placed in parallel with existing micro-OERs; Step 6: Use the prediction equation of preference value to evaluate the insertion MR n The predicted preference values ​​of all micro-OERs are shown below; Step 7: Apply heuristic rule 2 to re-evaluate the new learning path; Step 8: Use the hill climbing algorithm to calculate the fitness of the new learning path; Step 9: If the fitness of the new learning path is greater than the fitness of the existing learning path, the new learning path is used to replace the existing learning path; Step 10: Output the new learning path, which contains the newly inserted MR n learning path.

Images