Learning achievement degree estimation device, learning achievement degree estimation model construction device, method, and program
The learning achievement estimation device processes historical data to predict future learning behaviors and achievement levels, addressing the limitations of existing technologies by providing timely and accurate feedback for improving learning strategies across diverse subjects.
Patent Information
- Application Number
- PCT/JP2023/018448
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-17
- Publication Date
- 2024-11-21
AI Technical Summary
Current learning achievement estimation technologies are limited in their ability to predict final learning outcomes for learners across various subjects, especially those that do not use class materials, and fail to provide timely feedback on the effectiveness of learning methods, making it difficult for learners to adjust their strategies mid-stream.
A device and method that acquires and processes historical learning behavior data to generate chronological sequences, using neural network models to estimate future learning behaviors and achievement levels, allowing for the estimation of final learning achievement based on continued or changed learning methods.
Enables accurate prediction of final learning achievement levels and provides actionable feedback on improving learning behaviors, enabling learners to adjust their methods effectively.
Smart Images

Figure JP2023018448_21112024_PF_FP_ABST
Abstract
Description
Learning achievement estimation device, learning achievement estimation model construction device, method, and program
[0001] The disclosed technology relates to a learning achievement estimation device, a learning achievement estimation model construction device, a learning achievement estimation method, and a learning achievement estimation program.
[0002] Currently, technologies to support learners' learning have been proposed in areas such as schools, qualification exams, recurrent education, and reskilling. For example, a technology has been proposed that predicts student grades early, with the aim of reducing the number of students who fail courses. This technology uses a mathematical model to predict grades weekly based on students' browsing behavior regarding content such as class materials.
[0003] S. Leelaluk, T. Minematsu, Y. Taniguchi, F. Okubo, and A. Shimada, "Predicting student performance based on lecture materials data using neural network models," In Proceedings of the 4th Workshop on Predicting Performance Based on the Analysis of Reading Behavior, 2022.
[0004] It is important for learners to understand what learning methods will improve their final learning achievement, such as grades, in order to improve their learning achievement. Therefore, there is a need for a system that allows learners to realize midway through their studies what will happen to their final learning achievement if they continue with their current learning methods or if they change them.
[0005] The above-mentioned conventional technology is being considered for application to a limited number of subjects, but because it focuses on class materials, there is a possibility that it may not be applicable when the number of conditions for the type of subject increases, for example, to subjects that do not use class materials.
[0006] The disclosed technology has been made in consideration of the above points, and aims to estimate the final learning achievement level taking into account the continuation or change of learning method.
[0007] A first aspect of the present disclosure is a learning achievement estimation device, comprising: an acquisition unit that acquires information on a past behavior history and attributes related to learning of a target learner; a generation unit that generates first estimation data from the information acquired by the acquisition unit, the first estimation data including a behavior sequence in which the learning behavior of the target learner at multiple past time points is arranged in chronological order; and a first estimation model that, when the first estimation data for a specific learner is input, estimates the learning behavior of the specific learner at each of one or more future time points, by inputting the first estimation data for the target learner generated by the generation unit into a first estimation model that is pre-constructed so as to estimate the learning behavior of the specific learner at each of one or more future time points, The system includes a first estimation unit that estimates the learning behavior of the target learner at each of one or more future time points; a second estimation unit that inputs second estimation data including the behavioral sequence of the target learner's learning behavior at each of one or more future time points estimated by the first estimation unit into a second estimation model that is pre-constructed to estimate the learning achievement of the specific learner at a final time point when second estimation data including the behavioral sequence of the specific learner at multiple past and future time points is input, thereby estimating the learning achievement of the target learner at the final time point; and an output unit that outputs the learning achievement estimated by the second estimation unit.
[0008] A second aspect of the present disclosure is a learning achievement estimation model construction device including: an acquisition unit that acquires information on past learning behavior history, attributes, and learning achievement for each of a plurality of learners; a generation unit that generates training data from the information acquired by the acquisition unit, including a behavioral series in which the learning behavior of each of the plurality of learners at multiple past points in time is arranged in chronological order, and the learning achievement at a final point in time; a first construction unit that uses the training data generated by the generation unit to construct a first estimation model for estimating each learning behavior of the behavioral series included in the training data at one or more points in time included in a period immediately before the final point in time; and a second construction unit that uses the training data generated by the generation unit to construct a second estimation model for estimating the learning achievement at the final point in time included in the training data.
[0009] A third aspect of the present disclosure is a learning achievement estimation method executed by a learning achievement estimation device including an acquisition unit, a generation unit, a first estimation unit, a second estimation unit, and an output unit, wherein the acquisition unit acquires information on a past behavior history and attributes related to learning for a target learner, the generation unit generates first estimation data from the information acquired by the acquisition unit, the first estimation data including a behavior sequence in which the learning behavior of the target learner at multiple past time points is arranged in chronological order, and the first estimation unit, when the first estimation data for a specific learner is input, adds the generated data to a first estimation model that is pre-constructed to estimate the learning behavior of the specific learner at each of one or more future time points. the second estimation unit inputs the second estimation data including the behavioral sequence of the target learner's learning behavior at each of one or more future time points estimated by the first estimation unit into a second estimation model constructed in advance to estimate the learning achievement of the specific learner at a final time point when second estimation data including the behavioral sequence of the specific learner at multiple past and future time points is input, thereby estimating the learning achievement of the target learner at the final time point; and the output unit outputs the learning achievement estimated by the second estimation unit.
[0010] A fourth aspect of the present disclosure is a learning achievement estimation program, which is a program for causing a computer to function as each part of the learning achievement estimation device described above.
[0011] According to the disclosed technology, it is possible to estimate the final learning achievement level taking into account the continuation or change of the learning method.
[0012] 1 is a block diagram showing the hardware configuration of a learning achievement estimation model construction device and a learning achievement estimation device. FIG. 2 is a block diagram showing an example of the functional configuration of a learning achievement estimation model construction device. FIG. 3 is a diagram showing an example of a behavior sequence. FIG. 4 is a diagram showing an example of statistical information. FIG. 5 is a diagram showing an example of attribute information. FIG. 6 is a diagram explaining an example of construction of a learning behavior estimation model. FIG. 7 is a diagram explaining an example of construction of a learning achievement estimation model. FIG. 8 is a block diagram showing an example of the functional configuration of a learning achievement estimation device. FIG. 9 is a diagram explaining an example of estimation of learning achievement. FIG. 10 is a diagram explaining an example of estimation of learning behavior. FIG. 11 is a diagram explaining an example of estimation of learning achievement. FIG. 12 is a diagram explaining improvement of learning behavior. FIG. 13 is a flowchart showing the flow of a learning achievement estimation model construction process. FIG. 14 is a flowchart showing the flow of a learning achievement estimation process.
[0013] An example of an embodiment of the disclosed technology will be described below with reference to the drawings. In this embodiment, a case will be described in which a student's learning log is used to estimate the student's learning achievement, which is the student's final grade. A student is an example of a learner of the present disclosure. However, the technology of the present disclosure is not limited to this example and can also be applied to cases in which the learning achievement of working adults receiving training through e-learning, learners studying for qualification exams, etc.
[0014] In addition, the same reference numerals are used to denote the same or equivalent components and parts in each drawing. Also, the dimensional proportions of the drawings are exaggerated for the convenience of explanation and may differ from the actual proportions.
[0015] The learning achievement estimation system according to this embodiment includes a learning achievement estimation model construction device and a learning achievement estimation device.
[0016] Fig. 1 is a block diagram showing the hardware configuration of a learning achievement estimation model construction device 10. As shown in Fig. 1, the learning achievement estimation model construction device 10 has a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, a storage 14, an input unit 15, a display unit 16, and a communication I / F (Interface) 17. Each component is connected to each other via a bus 19 so as to be able to communicate with each other.
[0017] The CPU 11 is a central processing unit that executes various programs and controls each component. That is, the CPU 11 reads programs from the ROM 12 or the storage 14 and executes the programs using the RAM 13 as a work area. The CPU 11 controls the above components and performs various arithmetic processing in accordance with the programs stored in the ROM 12 or the storage 14. In this embodiment, the ROM 12 or the storage 14 stores a learning achievement estimation model construction program, which will be described later.
[0018] The ROM 12 stores various programs and various data. The RAM 13 temporarily stores programs or data as a working area. The storage 14 is configured by a storage device such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive), and stores various programs including an operating system and various data.
[0019] The input unit 15 includes a pointing device such as a mouse and a keyboard, and is used to input various types of information. The display unit 16 is, for example, a liquid crystal display, and displays various types of information. The display unit 16 may also function as the input unit 15 by employing a touch panel system.
[0020] The communication I / F 17 is an interface for communicating with other devices, and the communication may use a wired communication standard such as Ethernet (registered trademark) or FDDI, or a wireless communication standard such as 4G, 5G, or Wi-Fi (registered trademark).
[0021] The hardware configuration of the learning achievement estimation device 50 is similar to the hardware configuration of the learning achievement estimation model construction device 10 shown in Fig. 1, and therefore a description thereof will be omitted. In the case of the learning achievement estimation device 50, the ROM 12 or the storage 14 stores a learning achievement estimation program, which will be described later.
[0022] Next, the functional configuration of the learning achievement estimation model construction device 10 will be described. Fig. 2 is a block diagram showing an example of the functional configuration of the learning achievement estimation model construction device 10. Information on past behavioral history, attributes, and learning achievement related to learning for each of a plurality of students is input to the learning achievement estimation model construction device 10. In this embodiment, as shown in Fig. 2, information stored in each DB, namely, a learning behavior log DB (database) 30, a learning history DB 32, a subject information DB 34, and a student information DB 36, is input to the learning achievement estimation model construction device 10.
[0023] The learning behavior log DB 30 is a DB that stores logs such as the viewing history and operation history of electronic learning materials such as electronic textbooks and education management systems. The information stored in the learning behavior log DB 30 is an example of the above-mentioned past behavior history. The learning behavior log DB 30 stores information such as the learning materials viewed by the student and highlighted words in association with a student ID that is identification information for the student, a subject ID that is identification information for the subject, and date and time information for the viewing and operation.
[0024] The learning history DB 32 is a DB that stores information on various assignments and tests, course history, grades, and other learning results. The information stored in the learning history DB 32 is an example of the above-mentioned past behavioral history and learning achievement. The learning history DB 32 stores information such as test or assignment answers, scores, course content, and final grades, in association with student IDs, subject IDs, and date and time information such as the date and time of test or assignment administration and the date and time of attendance.
[0025] The subject information DB 34 is a DB that stores information about each subject. The information stored in the subject information DB 34 is information related to the past behavioral history and learning achievement level. The subject information DB 34 stores information such as the subject name, day and time of class, test questions, assignment content, information on the teaching materials used, and syllabus information for each subject, in association with the subject ID.
[0026] The student information DB 36 is a DB that stores information about each student. The information stored in the student information DB 36 is an example of the above-mentioned attributes. The student information DB 36 stores information such as each student's grade, alma mater, and entrance exam type in association with the student ID.
[0027] 2, the learning achievement estimation model construction device 10 includes, as functional components, a first acquisition unit 22, a first generation unit 24, a first construction unit 26, and a second construction unit 28. Each functional component is realized when the CPU 11 reads out a learning achievement estimation model construction program stored in the ROM 12 or the storage 14, expands it in the RAM 13, and executes it.
[0028] The first acquisition unit 22 acquires information on past learning behavior history, attributes, and learning achievement level for each of the multiple students. Specifically, the first acquisition unit 22 acquires information stored in each DB: the learning behavior log DB 30, the learning history DB 32, the subject information DB 34, and the student information DB 36. The first acquisition unit 22 may acquire all information stored in each DB, or may issue a query to each DB to acquire necessary information. The first acquisition unit 22 passes the acquired information to the first generation unit 24.
[0029] The first generation unit 24 generates training data from the information passed from the first acquisition unit 22, including a behavioral sequence in which the learning behaviors of each of multiple students at multiple points in time in the past are arranged in chronological order, and the learning achievement level at the final point in time.
[0030] Specifically, the first generation unit 24 extracts learning behavior for each student for each subject for each predetermined time period (e.g., every hour) from the learning behavior log DB 30 and the learning history DB 32. The first generation unit 24 generates a behavior sequence by chronologically arranging the extracted learning behavior, for example, as shown in FIG. 3 . In the example of FIG. 3 , the behavior sequence for each student is represented as information for each time period associated with each student's student ID and date and time information (e.g., 1 / 22 10:00-11:00). The information for each time period includes the start date and time of the time period (time), whether the date and time is a holiday (isHoliday), the subject ID (subject), and the learning behavior (action). The time, subject, and action are required fields, while the isHoliday field is optional.
[0031] In FIG. 3 , for example, "action:view,text1,10min" represents a learning behavior in which an electronic learning material identified by "text1" was viewed for 10 minutes. Also, for example, "action:test,test1,5pts" represents a score of 5 points on the test identified by "test1." Also, "action:eval,subjectT,B" represents the learning achievement level in the final time slot, and in this example, the learning achievement level for subject T is B. In this embodiment, the learning achievement levels are represented by S, A, B, C, D, etc., in descending order of achievement level. Furthermore, time slots in which no learning behavior occurred are padded with "PAD."
[0032] The first generating unit 24 may also include in the training data statistical information and attribute information obtained from the behavioral history and attributes, which are information passed from the first acquiring unit 22. Specifically, the first generating unit 24 generates statistical information including information extracted for each student by subject from each DB, and information obtained by combining or aggregating the extracted information. Figure 4 shows an example of the statistical information. In the example of Figure 4, the statistical information includes information such as student ID, subject ID, average test score, study timing, and learning achievement level.
[0033] For example, the first generation unit 24 calculates an average test score by averaging the scores for each test included in the information for each subject ID extracted for each student ID from the learning history DB 32. Furthermore, for example, the first generation unit 24 identifies, as study dates and times, the dates and times when each student viewed or operated electronic learning materials from the information in the learning behavior log DB 30, and the dates and times when each student performed assignments from the learning history DB 32. The first generation unit 24 then identifies the dates and times when tests or classes for each subject are held from the information in the subject information DB 34, and identifies study timings, such as the day before a test or the day before a class, based on the study dates and times relative to the dates and times of the tests or classes. Furthermore, for example, the first generation unit 24 identifies final learning achievement levels from final grades included in the learning history DB 32, and includes the identified learning achievement levels in the statistical information.
[0034] Furthermore, for example, the first generation unit 24 extracts information about each student for each student ID from the student information DB 36 and generates attribute information such as that shown in FIG. 5. In the example of FIG. 5, the attribute information includes information such as the student ID, academic year, alma mater, and GPA (Grade Point Average). The information included in the attribute information may directly include the information included in the student information DB 36, or may be information obtained by combining or aggregating information from the learning history DB 32, etc. For example, the first generation unit 24 may include an index (e.g., the GPA in FIG. 5) that comprehensively evaluates the test or assignment results of each student included in the learning history DB 32 in the attribute information.
[0035] The first generation unit 24 passes the generated statistical information, attribute information, and behavioral sequences to the first construction unit 26 and the second construction unit 28 as training data.
[0036] The first construction unit 26 constructs a learning behavior estimation model 40 for estimating each learning behavior at one or more time points included in a period immediately before the final time point, among the behavioral sequences included in the training data, using the training data passed from the first generation unit 24. The learning behavior estimation model 40 is an example of a first estimation model of the present disclosure.
[0037] Specifically, the first construction unit 26 masks each of the learned behaviors at one or more time points included in the period immediately prior to the final time point, from among the behavioral sequences included in the training data. The number of learned behaviors to be masked may be set appropriately depending on the number of future learned behaviors to be estimated. The first construction unit 26 then constructs the learned behavior estimation model 40 by updating the parameters of the learned behavior estimation model 40 so that the estimation result of the masked portion by the learned behavior estimation model 40 matches the learned behavior before masking.
[0038] For example, the Masked Language Model used in BERT (Bidirectional Encoder Representations from Transformers) can be used as the learning behavior estimation model 40. In this case, as shown in the upper diagram of FIG. 6 , the first constructor 26 masks ([MASK]) learning behaviors (actions) in a predetermined number of time periods in the latter half of the time series in the behavior sequence. Then, as shown in FIG. 7 , the first constructor 26 inputs date and time information indicating the first time period, the learning behavior of the behavior sequence in that time period, statistical information, and attribute information into BERT as vector sequence 1 in the order of time periods in the behavior sequence. The first constructor 26 sequentially inputs information in the time series up to the last time period in the behavior sequence. At this time, the first constructor 26 generates a vector sequence by combining the embedded information.
[0039] For example, the embedding of time information may be performed using a periodic function. Furthermore, the embedding of learning behavior is initialized for each learning behavior with the number of dimensions of the hidden layer of BERT, and the number is updated each time machine learning is performed. Each learning behavior is assigned an identifier in advance in a dictionary, and calculations are performed using the identifier. The dictionary also includes special identifiers in advance: an identifier [UNK] indicating an unknown learning behavior, an identifier [CLS] indicating the start of a behavior sequence, an identifier [SEP] indicating the end of a behavior sequence, and an identifier [PAD] used to fill in when the sequence does not meet a predetermined length. The embedding of learning behavior is not limited to this, and may also be a transformation in which an identifier indicating one learning behavior, called a one-hot vector, is assigned to one dimension.
[0040] Furthermore, for example, the embedding of numerical values used in statistical information and attribute information is initialized with the number of the dimension of the hidden layer of BERT for each numerical sequence, and the number is updated each time machine learning is performed. In other words, if the numerical value before embedding is x (a scalar) and the embedding vector associated with the sequence is W (a vector of the dimension of the hidden layer), then x × W is the value after embedding, and W is updated by machine learning.
[0041] As shown in the lower diagram of FIG. 6 , when an estimation result of the masked portion is obtained, the first construction unit 26 repeatedly updates the parameters of the learning behavior estimation model 40 so that the estimation result matches the information before masking, until an end condition for machine learning is met. The end condition for machine learning may be when the parameter update is repeated a predetermined number of times, when the error between the estimation result and the information before masking becomes equal to or less than a predetermined value, or when the difference between the previous error and the current error becomes equal to or less than a predetermined value, i.e., when the error converges. The first construction unit 26 outputs the learning behavior estimation model 40 in which the parameters are set when the end condition for machine learning is met. The output learning behavior estimation model 40 is stored in a predetermined storage area of the learning achievement estimation device 50.
[0042] The first construction unit 26 may construct the learning behavior estimation model 40 by applying an algorithm other than BERT. For example, it is possible to construct a model that estimates learning behaviors one by one using a classical machine learning algorithm. If the influence of learning behaviors is not significantly reflected in the estimation results, it is possible to increase the influence of learning behaviors by dividing the input information into multiple attention mechanisms and strengthening attention to learning behaviors.
[0043] The second construction unit 28 constructs a learning achievement estimation model 42 for estimating the learning achievement at the final point in time included in the training data, using the training data passed from the first generation unit 24. The learning achievement estimation model 42 is an example of a second estimation model of the present disclosure.
[0044] Specifically, the second construction unit 28 masks the learning achievement at the final time point of the behavioral sequence included in the training data. Then, the second construction unit 28 constructs the learning achievement estimation model 42 by updating the parameters of the learning achievement estimation model 42 so that the estimation result of the masked part by the learning achievement estimation model 42 matches the learning achievement before masking.
[0045] For example, the learning achievement estimation model 42, like the learning behavior estimation model 40, can also use the Masked Language Model used in BERT. In this case, as shown in the upper diagram of FIG. 8 , the second construction unit 28 masks ([MASK]) the learning behavior (action) for the final time period, i.e., the learning achievement portion. Similarly to the first construction unit 26, the second construction unit 28 inputs date and time information indicating the first time period, the learning behavior of the behavior series for that time period, statistical information, and attribute information into BERT in the order of time periods in the behavior series as Vector Sequence 1. The second construction unit 28 sequentially inputs information in chronological order up to the last time period of the behavior series. Then, as shown in the lower diagram of FIG. 8 , when an estimation result for the masked portion is obtained, the second construction unit 28 repeatedly updates the parameters of the learning achievement estimation model 42 until the machine learning termination condition is met so that the estimation result matches the information before masking. The second construction unit 28 outputs the learning achievement estimation model 42 in which the parameters set when the machine learning termination condition is satisfied. The output learning achievement estimation model 42 is stored in a predetermined storage area of the learning achievement estimation device 50.
[0046] The second constructor 28 may apply an algorithm other than BERT to construct the learning achievement estimation model 42. For example, it is also possible to estimate the learning achievement using a classical machine learning algorithm.
[0047] In this manner, in this embodiment, two models, the learning behavior estimation model 40 and the learning achievement estimation model 42, are constructed by using the same training data and changing the masked portions.
[0048] Next, a functional configuration of the learning achievement estimation device 50 will be described. Fig. 9 is a block diagram showing an example of the functional configuration of the learning achievement estimation device 50. As shown in Fig. 9, the learning achievement estimation device 50 includes, as functional components, a second acquisition unit 52, a second generation unit 54, a first estimation unit 56, a second estimation unit 58, and an output unit 60. Each functional component is realized when the CPU 11 reads out a learning achievement estimation program stored in the ROM 12 or the storage 14, expands it in the RAM 13, and executes it.
[0049] The second acquisition unit 52 acquires information on the past learning behavior history and attributes of a target student, whose learning achievement level is to be estimated. Specifically, the second acquisition unit 52 acquires information associated with the student ID of the target student from each of the learning behavior log DB 30, the learning history DB 32, the subject information DB 34, and the student information DB 36. The second acquisition unit 52 passes the acquired information to the second generation unit 54.
[0050] The second generating unit 54 generates first estimation data including the target student's behavioral sequence at multiple past points in time from the information received from the second acquiring unit 52. The first estimation data may include statistical information and attribute information, similar to the training data. The method for generating the statistical information and attribute information is the same as the method for generating the statistical information and attribute information by the first generating unit 24. Similarly to the first generating unit 24, the second generating unit 54 generates a behavioral sequence as shown in FIG. 3 . However, the second generating unit 54 generates a behavioral sequence from the present to the past based on the learning behavior log DB 30 and the learning history DB 32, and also generates a future behavioral sequence for a time period for which learning behavior and learning achievement are desired to be estimated, and combines these. The future behavioral sequence includes a future date and time (time), a subject ID (subject), and learning behavior (action). Furthermore, an optional item indicating whether the date and time is a holiday (isHoliday) may be included. As shown in the upper diagram of Fig. 10, the learning action of the future action sequence is an identifier indicating that it is the learning action to be estimated (indicated as "[future action]" in the example of Fig. 10). Also, the learning action of the final time slot is an identifier indicating that it is the learning achievement level to be estimated (indicated as "[TARGET]" in the example of Fig. 10).
[0051] The second generation unit 54 passes the generated statistical information, attribute information, and behavioral sequence about the target student to the first estimation unit 56 and the second estimation unit 58 as first estimation data.
[0052] The first estimation unit 56 estimates learning behavior for a future time period in the action sequence of the first estimation data by inputting the first estimation data passed from the second generation unit 54 into the learning behavior estimation model 40 constructed by the learning achievement estimation model construction device 10. For example, as shown in the dashed line portion in the middle diagram of Fig. 10 , the "[future action]" portion of the learning behavior (action) for a future time period in the action sequence is estimated.
[0053] More specifically, as shown in Fig. 11 , the first estimation unit 56 inputs first estimation data consisting of statistical information, attribute information, and a behavioral sequence including a future time period about a target student (Student X in the example of Fig. 11 ) to the learning behavior estimation model 40. The learning behavior estimation model 40 estimates the probability that the learning behavior in the time period to be estimated is each of predetermined learning behaviors (in the example of Fig. 11 , behavior a, behavior b, behavior c, ..., behavior y, and behavior z), and outputs the behavior with the highest probability as the estimation result. The example of Fig. 11 shows an example in which the learning behavior in the time period of date and time information "11 / 1 20:00-21:00" is estimated to be "behavior c," and the learning behavior in the time period of date and time information "11 / 2 0:00-1:00" is estimated to be "behavior y."
[0054] The first estimation unit 56 passes the estimation result to the second estimation unit 58 .
[0055] The second estimation unit 58 generates second estimation data by applying the estimation result passed from the first estimation unit 56 to the first estimation data passed from the second generation unit 54. That is, the second estimation data includes a behavioral sequence in which future learning behavior has been estimated. The second estimation unit 58 inputs the generated second estimation data into the learning achievement estimation model 42 constructed by the learning achievement estimation model construction device 10, thereby estimating the learning behavior in the final time period of the behavioral sequence of the second estimation data, i.e., the learning achievement. For example, as shown by the dashed-dotted line in the lower diagram of FIG. 10 , the "[TARGET]" portion of the learning behavior (action) in the final time period of the behavioral sequence is estimated.
[0056] More specifically, as shown in Fig. 12, the second estimation unit 58 inputs second estimation data consisting of statistical information, attribute information, and a behavioral sequence including estimated future learning behavior about a target student (Student X in the example of Fig. 12) into the learning achievement estimation model 42. The learning achievement estimation model 42 estimates the probability that the learning achievement corresponding to the [TARGET] portion is each of predetermined learning achievement levels (in the example of Fig. 12, grade S, grade A, ..., grade D), and outputs the learning achievement level with the highest probability as the estimation result. The example of Fig. 12 shows that the time period with date and time information "11 / 2 4:00-5:00" is the final time period, and the learning achievement level for that time period is estimated to be "Grade D."
[0057] The second estimation unit 58 passes the estimation result to the output unit 60 .
[0058] If the learning achievement level estimated by the second estimating unit 58 has not reached a predetermined target value, the output unit 60 replaces at least one of the learning behaviors at each of one or more future time points estimated by the first estimating unit 56 with a learning behavior that is improved in a direction that improves the learning achievement level. Then, the output unit 60 causes the second estimating unit 58 to estimate the learning achievement level again using the second estimation data including the behavior sequence after the replacement.
[0059] Specifically, the output unit 60 calculates the contribution of each learning behavior estimated by the first estimation unit 56 to the estimation result of the learning achievement by the second estimation unit 58. The output unit 60 generates, as an improvement plan, a learning behavior that improves the learning achievement by improving the learning behavior whose contribution is equal to or greater than a predetermined value or whose contribution is in the top predetermined range. For example, the output unit 60 may predefine rules for generating improvement plans and generate the improvement plans using these rules. Alternatively, the output unit 60 may pre-construct a model for outputting improvement plans using machine learning and generate the improvement plans using this model. The output unit 60 overwrites the original learning behavior with the generated improvement plan.
[0060] For example, as shown in FIG. 13 , assume that the learning achievement level estimated by the second estimating unit 58 is "D" and the predetermined target value is "B." In this case, the output unit 60 calculates the contribution of learning behaviors for each time period of date and time information "2023 / 3 / 6 15:00-16:00" and "2023 / 3 / 6 16:00-17:00," which are the estimation results of the first estimating unit 56, to the estimation result of the learning achievement level. The output unit 60 calculates the contribution using, for example, the strength of attention, a SHAP (Shapley Additive explanations) value, etc.
[0061] In the example of FIG. 13 , the second estimation unit 58 generates an improvement plan for learning behavior during the time period of the date and time information "2023 / 3 / 6 15:00-16:00" with a high contribution. In the example of FIG. 13 , the corresponding learning behavior before improvement indicates that the score on the test identified by "test1" was 30 points. Therefore, the output unit 60 generates, for example, an improvement plan for learning behavior indicating that the score on the test identified by "test1" is 80 points. This is an improvement plan that represents improving the learning behavior so that 80 points can be obtained on the test identified by "test1" at this point. Note that the score after improvement is a score derived, for example, by a rule or a machine learning model. Furthermore, the improvement plan is not limited to the above example. It may also be an improvement plan that changes the type of learning behavior, such as when the learning behavior before improvement indicates viewing electronic learning materials and the improvement plan is learning behavior indicating completing an assignment.
[0062] If the second estimation unit 58 estimates the learning achievement level using the improved second estimation data to be "B," the output unit 60 outputs the estimation result together with an improvement proposal for when the estimation result is obtained. The output unit 60 may output the improvement proposal as a message such as, for example, "If your score on Test 1 remains the same, it will be 30 points. In order to achieve learning achievement level B, please improve your learning behavior so that your score on Test 1 will be 80 points."
[0063] If the initial estimation result of the learning achievement level reaches the target value, no improvement plan is generated, and therefore the output unit 60 only needs to output the estimation result of the learning achievement level. Also, the output unit 60 may output the initial estimation result of the learning achievement level and the improvement plan without re-estimating the learning achievement level using the second estimation data after replacement with the improvement plan. Also, output of the improvement plan is not essential, and the output unit 60 may output the initial estimation results of the learning behavior and the learning achievement level as they are.
[0064] Next, a description will be given of the operation of the learning achievement estimation model construction device 10. Fig. 14 is a flowchart showing the flow of the learning achievement estimation model construction process by the learning achievement estimation model construction device 10. The learning achievement estimation model construction process is performed by the CPU 11 reading out a learning achievement estimation model construction program from the ROM 12 or storage 14, expanding it into the RAM 13, and executing it.
[0065] In step S10, the CPU 11, as the first acquisition unit 22, acquires information stored in each of the DBs, namely, the learning behavior log DB 30, the learning history DB 32, the subject information DB 34, and the student information DB 36. Then, the CPU 11, as the first generation unit 24, generates statistical information including information extracted for each subject for each student from each DB and information obtained by combining or aggregating the extracted information. The CPU 11, as the first generation unit 24, also extracts information about each student from the student information DB 36 and generates attribute information for each student based on the extracted information.
[0066] Next, in step S12, the CPU 11, as the first generation unit 24, extracts learning behaviors for each subject and for each predetermined time period (e.g., every hour) for each student from the learning behavior log DB 30 and the learning history DB 32. Then, the CPU 11, as the first generation unit 24, arranges the extracted learning behaviors in chronological order to generate a behavior sequence.
[0067] Next, in step S14, the CPU 11, as the first construction unit 26, masks each of the learned behaviors of the behavior series at one or more time points included in the period immediately before the final time point in the training data consisting of statistical information, attribute information, and behavior series. Then, the CPU 11, as the first construction unit 26, constructs the learned behavior estimation model 40 by updating the parameters of the learned behavior estimation model 40 so that the estimation result of the masked portion by the learned behavior estimation model 40 matches the learned behavior before masking. The CPU 11, as the first construction unit 26, outputs the constructed learned behavior estimation model 40.
[0068] Next, in step S16, the CPU 11, functioning as the second construction unit 28, masks the learning achievement at the final point in the behavioral sequence in the training data consisting of statistical information, attribute information, and behavioral sequences. Then, the CPU 11, functioning as the second construction unit 28, constructs the learning achievement estimation model 42 by updating the parameters of the learning achievement estimation model 42 so that the estimation result of the masked portion by the learning achievement estimation model 42 matches the learning achievement before masking. The CPU 11, functioning as the second construction unit 28, stores the constructed learning achievement estimation model 42 in a predetermined storage area, and the learning achievement estimation model construction process ends.
[0069] Next, a description will be given of the operation of the learning achievement estimation device 50. Fig. 15 is a flowchart showing the flow of the learning achievement estimation process performed by the learning achievement estimation device 50. The learning achievement estimation process is performed by the CPU 11 reading out a learning achievement estimation program from the ROM 12 or the storage 14, expanding it into the RAM 13, and executing it. The learning achievement estimation process is an example of a learning achievement estimation method disclosed herein.
[0070] In step S20, the CPU 11, as the second acquisition unit 52, acquires information associated with the student ID of the target student from each of the learning behavior log DB 30, learning history DB 32, subject information DB 34, and student information DB 36. Then, the CPU 11, as the second generation unit 54, generates statistical information from the acquired information, including information extracted for each subject about the target student and information obtained by combining or aggregating the extracted information. Furthermore, the CPU 11, as the second generation unit 24, generates attribute information of the target student based on the information of the target student acquired from the student information DB 36.
[0071] Next, in step S22, the CPU 11, as the second generation unit 54, extracts the learning behavior of the target student for each subject and for each predetermined time period (e.g., every hour) from the information acquired from the learning behavior log DB 30 and the learning history DB 32. Then, the CPU 11, as the second generation unit 54, generates a behavior series by combining a behavior series from the present to the past in which the extracted learning behaviors are arranged in chronological order with a future behavior series for a time period in which the learning behavior and learning achievement level are desired to be estimated.
[0072] Next, in step S24, the CPU 11, functioning as the first estimation unit 56, inputs first estimation data consisting of statistical information, attribute information, and behavioral sequences into the learning behavior estimation model 40 constructed by the learning achievement estimation model construction device 10. As a result, the CPU 11, functioning as the first estimation unit 56, estimates learning behavior for a future time period of the behavioral sequences of the first estimation data.
[0073] Next, in step S26, the CPU 11, functioning as the second estimation unit 58, generates second estimation data by applying the estimation result passed from the first estimation unit 56 to the first estimation data. The CPU 11, functioning as the second estimation unit 58, inputs the generated second estimation data into the learning achievement estimation model 42 constructed by the learning achievement estimation model construction device 10, thereby estimating the learning behavior, i.e., the learning achievement, for the final time period of the behavior sequence of the second estimation data.
[0074] Next, in step S28, the CPU 11, functioning as the output unit 60, determines whether or not the learning achievement level estimated in step S26 has reached a predetermined target value. If the target value has been reached, the process proceeds to step S32, and if not, the process proceeds to step S30.
[0075] In step S30, the CPU 11, as the output unit 60, calculates the contribution of each of the learning behaviors estimated in step S24 to the estimation result in step S28. The CPU 11, as the output unit 60, generates an improvement plan for the learning behaviors whose contribution is equal to or greater than a predetermined value or whose contribution is among the top predetermined numbers of learning behaviors, improving the learning achievement level. The CPU 11, as the output unit 60, then causes the second estimation unit 58 to re-estimate the learning achievement level using the second estimation data in which the original learning behavior has been overwritten with the generated improvement plan, and returns to step S28.
[0076] In step S32, the CPU 11 outputs, as the output unit 60, the learning achievement level when the target value is reached and, if an improvement plan has been generated in step S30, the improvement plan, and the learning achievement level estimation process ends.
[0077] As described above, the learning achievement estimation device according to this embodiment acquires information on the past learning behavior history and attributes of a target learner. Furthermore, the learning achievement estimation device generates, from the acquired information, first estimation data including a behavioral sequence in which the target learner's learning behavior at multiple past time points is arranged in chronological order. The learning achievement estimation device then inputs the first estimation data about the target learner into a learning behavior estimation model to estimate the learning behavior of the target learner at each of one or more future time points. Furthermore, the learning achievement estimation device inputs second estimation data including a behavioral sequence of the learning behavior estimated for the target learner at each of one or more future time points into the learning achievement estimation model to estimate and output the target learner's learning achievement at a final time point. The learning behavior estimation model is pre-constructed by the learning achievement estimation model construction device so as to estimate the learning behavior of a specific learner at each of one or more future time points when the first estimation data about the specific learner is input. In addition, the learning achievement estimation model is pre-constructed by the learning achievement estimation model construction device so that when second estimation data including a behavioral sequence of a specific learner at multiple points in the past and future is input, the learning achievement estimation model is capable of estimating the learning achievement of a specific learner at the final point in time.
[0078] In this way, the learning achievement estimation device according to the present embodiment estimates the future learning behavior of the target learner and then estimates the learning achievement if that learning behavior is performed, and therefore can estimate the final learning achievement taking into account the continuation or change of the learning method. In addition, the learning achievement estimation device according to the present embodiment can propose future learning behavior as an improvement plan that will improve the final learning achievement.
[0079] In the above embodiment, the processing performed by the CPU after reading the software (program) may be performed by various processors other than the CPU. Examples of processors in this case include a PLD (Programmable Logic Device) whose circuit configuration can be changed after manufacture, such as an FPGA (Field-Programmable Gate Array), and a dedicated electrical circuit, such as an ASIC (Application Specific Integrated Circuit), which is a processor having a circuit configuration designed specifically to perform specific processing. Furthermore, the above processing may be performed by one of these various processors, or by a combination of two or more processors of the same or different types (e.g., multiple FPGAs, or a combination of a CPU and an FPGA). Furthermore, the hardware structure of these various processors is, more specifically, an electrical circuit that combines circuit elements such as semiconductor elements.
[0080] In the above embodiment, the program is pre-stored (installed) in a ROM or storage device, but the present invention is not limited to this. The program may be provided in a form stored on a non-transitory storage medium such as a CD-ROM (Compact Disk Read Only Memory), a DVD-ROM (Digital Versatile Disk Read Only Memory), or a USB (Universal Serial Bus) memory. The program may also be downloaded from an external device via a network.
[0081] The following additional notes are provided regarding the above-described embodiments.
[0082] (Supplementary Item 1) A learning achievement estimation device comprising: a memory; and at least one processor connected to the memory, wherein the processor is configured to: acquire information on past learning behavior history and attributes of a target learner; generate, from the acquired information, first estimation data including a behavioral sequence in which the learning behavior of the target learner at multiple past time points is arranged in chronological order; when the first estimation data for a specific learner is input, input the generated first estimation data for the target learner into a first estimation model that is pre-constructed to estimate the learning behavior of the specific learner at each of one or more future time points, thereby estimating the learning behavior of the target learner at each of one or more future time points; when second estimation data including the behavioral sequence of the specific learner at multiple past and future time points is input, input second estimation data including the estimated behavioral sequence of the target learner's learning behavior at each of one or more future time points into a second estimation model that is pre-constructed to estimate the learning achievement of the specific learner at a final time point; and output the estimated learning achievement.
[0083] (Supplementary Item 2) A learning achievement estimation model construction device including: a memory; and at least one processor connected to the memory, wherein the processor is configured to: acquire information on past learning-related behavioral history, attributes, and learning achievement of each of a plurality of learners; generate, from the acquired information, training data including an action sequence in which the learning behavior of each of the plurality of learners at multiple past time points is arranged in chronological order, and the learning achievement at a final time point; use the generated training data to construct a first estimation model for estimating each learning behavior of the action sequence included in the training data at one or more time points included in a period immediately before the final time point; and use the generated training data to construct a second estimation model for estimating the learning achievement at the final time point included in the training data.
[0084] (Supplementary clause 3) A non-transitory storage medium storing a program executable by a computer to execute a learning achievement estimation process, wherein the learning achievement estimation process: acquires information on past behavioral history and attributes related to learning for a target learner; generates first estimation data from the acquired information, including a behavioral sequence in which the learning behavior of the target learner at multiple past time points is arranged in chronological order; inputs the generated first estimation data for the target learner into a first estimation model that is pre-constructed to estimate the learning behavior of the specific learner at each of one or more future time points when the first estimation data for the specific learner is input, thereby estimating the learning behavior of the target learner at each of one or more future time points; inputs second estimation data including the behavioral sequence of the specific learner at multiple past and future time points when inputs second estimation data including the estimated behavioral sequence of the target learner's learning behavior at each of one or more future time points into a second estimation model that is pre-constructed to estimate the learning achievement of the specific learner at a final time point, thereby estimating the learning achievement of the target learner at the final time point; and outputs the estimated learning achievement. Non-transitory storage medium.
[0085] (Supplementary Item 4) A non-transitory storage medium storing a program executable by a computer to execute a learning achievement estimation model construction process, wherein the learning achievement estimation model construction process: acquires information on past learning behavior history, attributes, and learning achievement for each of a plurality of learners; generates, from the acquired information, training data including a behavioral sequence in which the learning behaviors of each of the plurality of learners at multiple past time points are arranged in chronological order, and the learning achievement at a final time point; uses the generated training data to construct a first estimation model for estimating each learning behavior of the behavioral sequence included in the training data at one or more time points included in a period immediately before the final time point; and uses the generated training data to construct a second estimation model for estimating the learning achievement at the final time point included in the training data.
[0086] REFERENCE SIGNS LIST 10 Learning achievement estimation model construction device 11 CPU 12 ROM 13 RAM 14 Storage 15 Input unit 16 Display unit 17 Communication I / F 19 Bus 22 First acquisition unit 24 First generation unit 26 First construction unit 28 Second construction unit 30 Learning behavior log DB 32 Learning history DB 34 Subject information DB 36 Student information DB 40 Learning behavior estimation model 42 Learning achievement estimation model 50 Learning achievement estimation device 52 Second acquisition unit 54 Second generation unit 56 First estimation unit 58 Second estimation unit 60 Output unit
Claims
1. An acquisition unit that acquires information on past behavioral history and attributes related to learning for a target learner; a generation unit that generates first estimation data including a behavioral sequence in which the learning behavior of the target learner at multiple past time points is arranged in chronological order from the information acquired by the acquisition unit; a first estimation unit that, when the first estimation data for a specific learner is input, inputs the first estimation data for the target learner generated by the generation unit into a first estimation model that is pre-constructed to estimate the learning behavior of the specific learner at each of one or more future time points, and estimates the learning behavior of the target learner at each of one or more future time points; a second estimation unit that, when second estimation data including the behavioral sequence of the specific learner at multiple past and future time points is input, inputs second estimation data including the behavioral sequence of the learning behavior of the target learner at each of one or more future time points estimated by the first estimation unit into a second estimation model that is pre-constructed to estimate the learning achievement of the specific learner at a final time point, and estimates the learning achievement of the target learner at a final time point; and an output unit that outputs the learning achievement estimated by the second estimation unit. A learning achievement estimation device comprising:
2. A learning achievement estimation device as described in claim 1, wherein the generation unit generates the first estimation data including statistical information and attribute information obtained from the behavioral history and the attributes.
3. A learning achievement estimation device as described in claim 1 or claim 2, wherein the output unit outputs the contribution of each learning behavior at each of one or more future points in time estimated by the first estimation unit to the learning achievement estimation result by the second estimation unit.
4. A learning achievement estimation device as described in claim 3, wherein the output unit outputs as an improvement proposal a learning behavior that improves the learning achievement level, the learning behavior having a contribution level equal to or above a predetermined value or a top predetermined number of contribution levels.
5. A learning achievement estimation device as described in claim 1 or claim 2, wherein, when the learning achievement estimated by the second estimation unit has not reached a predetermined target value, the output unit replaces at least one of the learning actions at each of one or more future time points estimated by the first estimation unit with an improved learning action in a direction that increases the learning achievement, causes the second estimation unit to estimate the learning achievement again, and outputs the improved learning action as an improvement proposal when the learning achievement reaches the target value.
6. A learning achievement estimation model construction device including: an acquisition unit that acquires information on past learning-related behavior history, attributes, and learning achievement for each of a plurality of learners; a generation unit that generates training data from the information acquired by the acquisition unit, the training data including a behavior series in which the learning behavior of each of the plurality of learners at multiple past points in time is arranged in chronological order, and the learning achievement at a final point in time; a first construction unit that uses the training data generated by the generation unit to construct a first estimation model for estimating each learning behavior of the behavior series included in the training data at one or more points in time included in a period immediately before the final point in time; and a second construction unit that uses the training data generated by the generation unit to construct a second estimation model for estimating the learning achievement at the final point in time included in the training data.
7. A learning achievement estimation method executed by a learning achievement estimation device including an acquisition unit, a generation unit, a first estimation unit, a second estimation unit, and an output unit, wherein the acquisition unit acquires information on past behavioral history and attributes related to learning for a target learner, the generation unit generates first estimation data including a behavior sequence in which the learning behavior of the target learner at multiple past points in time is arranged in chronological order from the information acquired by the acquisition unit, and the first estimation unit inputs the first estimation data for the target learner generated by the generation unit into a first estimation model that is pre-constructed to estimate the learning behavior of the specific learner at each of one or more future points in time when the first estimation data for a specific learner is input, and estimates the learning behavior of the target learner at each of one or more future points in time, A learning achievement estimation method, in which the second estimation unit inputs second estimation data including a behavioral sequence of the learning behavior of the target learner at each of one or more future time points estimated by the first estimation unit into a second estimation model pre-constructed to estimate the learning achievement of the specific learner at a final time point when second estimation data including a behavioral sequence of the specific learner at multiple past and future time points is input, thereby estimating the learning achievement of the target learner at the final time point, and the output unit outputs the learning achievement estimated by the second estimation unit.
8. A learning achievement estimation program for causing a computer to function as each part of the learning achievement estimation device according to claim 1 or 2.
Citation Information
Patent Citations
Learning support device, learning support system, learning support method, and program
JP2005055659A
Device, method, and program for processing information
JP2018189726A
Evaluation program and device
JP2023027458A
Learning material recommendation method, learning material recommendation device, and learning material recommendation program
WO2018168220A1
Cited By
Time discount rate estimation apparatus, machine learning method, time discount rate analysis method, and program
US20250111387A1
