JP7604991B2 - Information processing device, information processing method, and program - Google Patents

Description

The present invention relates to an information processing device, an information processing method, and a program.

In recent years, systems to provide learning support to students have been developed. For example, Patent Document 1 discloses a group education method using a computer system that allows a teacher or corrector to monitor the progress of students' answers to assignments and provide comments on the answers.

JP 2015-018096 A

However, the invention described in Patent Document 1 cannot check the validity of the comments made by an instructor (e.g., a teacher or corrector).

In one aspect, the objective is to provide an information processing device etc. capable of checking the validity of an instructor's comments.

An information processing device according to one aspect includes a reception unit that receives time-series learning status data indicating the learning status of a learner and an instructor's comments on the learning status data, and an output unit that inputs the learning status data and comments received by the reception unit into a first learning model that has been trained to estimate the validity of the comments when time-series learning status data for a predetermined period and comments on the learning status data are input, and outputs an estimation result that estimates the validity of the comments.

On one hand, it allows for checking the validity of the instructor's comments.

FIG. 1 is an explanatory diagram showing an overview of a comment check system. FIG. 2 is a block diagram showing an example of the configuration of a server. FIG. 2 is an explanatory diagram illustrating an example of a record layout of a training data DB. 1 is an explanatory diagram showing an example of a record layout of a learning model management DB and a learner DB. FIG. 13 is an explanatory diagram illustrating an example of a record layout of a comment DB. FIG. 2 is a block diagram showing an example of the configuration of a learner terminal. 3 is a block diagram showing a configuration example of an instructor terminal; FIG. FIG. 13 is an explanatory diagram of a validity estimation model. 13 is a flowchart showing a procedure for generating a validity estimation model. FIG. 13 is an explanatory diagram showing an example of a display screen of an estimation result of the validity of a comment. 13 is a flowchart showing a processing procedure for performing a process of estimating the validity of a comment. 13 is a flowchart showing a processing procedure for performing a process of estimating the validity of a comment. FIG. 11 is a block diagram showing an example of the configuration of a server according to a second embodiment. FIG. 13 is an explanatory diagram showing an example of a record layout of a training data DB according to the second embodiment. FIG. 1 is an explanatory diagram regarding a comment estimation model. FIG. 13 is an explanatory diagram showing an example of a display screen of a comment estimation result. 13 is a flowchart showing a processing procedure for performing a comment generation process.

The present invention will be described in detail below with reference to drawings showing an embodiment of the invention.

(Embodiment 1)
The first embodiment relates to a form in which the validity of an instructor's comments is estimated using artificial intelligence (AI) based on learning situation data indicating a learner's learning situation and the instructor's comments on the learning situation data. Fig. 1 is an explanatory diagram showing an overview of a comment check system. The system of this embodiment includes an information processing device 1, an information processing terminal 2, and an information processing terminal 3, and each device transmits and receives information via a network N such as the Internet.

The information processing device 1 is an information processing device that processes, stores, and transmits/receives various types of information. The information processing device 1 is, for example, a server device, a personal computer, or a general-purpose tablet PC (personal computer). In this embodiment, the information processing device 1 is a server device, and for the sake of simplicity, will be referred to as server 1 below.

The information processing terminal 2 is a terminal device that receives and transmits the learner's learning status data, and receives and displays the instructor's comments. The information processing terminal 2 is, for example, an information processing device such as a smartphone, a mobile phone, a wearable device such as an Apple Watch (registered trademark), a tablet, or a personal computer terminal. For the sake of simplicity, the information processing terminal 2 will be referred to as the learner's terminal 2 below.

The information processing terminal 3 is a terminal device that receives and transmits comments from an instructor on the learning situation data of a learner, and receives and displays the estimation results of the validity of the comments. The information processing terminal 3 is, for example, an information processing device such as a smartphone, a mobile phone, a wearable device such as an Apple Watch, a tablet, or a personal computer terminal. For the sake of simplicity, the information processing terminal 3 will be referred to as the instructor terminal 3 below.

The server 1 according to this embodiment acquires time-series learning status data indicating the learning status of the learner, and the instructor's comments on the learning status data. The server 1 inputs the acquired learning status data and the instructor's comments into a learning model that has been trained to estimate the validity of the comments when time-series learning status data for a predetermined period and comments on the learning status data are input, and outputs an estimation result estimating the validity of the comments. The learning model will be described later.

Figure 2 is a block diagram showing an example of the configuration of the server 1. The server 1 includes a control unit 11, a storage unit 12, a communication unit 13, an input unit 14, a display unit 15, a reading unit 16, and a large-capacity storage unit 17. Each component is connected by a bus B.

The control unit 11 includes a processor such as a CPU (Central Processing Unit), MPU (Micro-Processing Unit), and GPU (Graphics Processing Unit), and performs various information processing, control processing, and the like related to the server 1 by reading and executing the control program 1P stored in the storage unit 12. The control program 1P can be deployed so that it is executed on a single computer or at one site, or distributed across multiple sites and on multiple computers interconnected by a communications network. Although the control unit 11 is described in FIG. 2 as being a single processor, it may also be a multi-processor.

The storage unit 12 includes memory elements such as RAM (Random Access Memory) and ROM (Read Only Memory), and stores the control program 1P or data required for the control unit 11 to execute processing. The storage unit 12 also temporarily stores data required for the control unit 11 to execute arithmetic processing. The communication unit 13 is a communication module for performing communication-related processing, and transmits and receives information between the learner terminal 2 and the instructor terminal 3 via the network N.

The input unit 14 is an input device such as a mouse, keyboard, touch panel, or button, and outputs received operation information to the control unit 11. The display unit 15 is a liquid crystal display or an organic EL (electroluminescence) display, etc., and displays various information according to instructions from the control unit 11.

The reading unit 16 reads the portable storage medium 1a, which may be a CD (Compact Disc)-ROM or a DVD (Digital Versatile Disc)-ROM. The control unit 11 may read the control program 1P from the portable storage medium 1a via the reading unit 16 and store it in the mass storage unit 17. The control unit 11 may also download the control program 1P from another computer via a network N or the like and store it in the mass storage unit 17. The control unit 11 may also read the control program 1P from the semiconductor memory 1b.

The large-capacity storage unit 17 includes a recording medium such as a hard disk drive (HDD) or a solid state drive (SSD). The large-capacity storage unit 17 includes a validity estimation model (first learning model) 171, a training data DB (database) 172, a learning model management DB 173, a learner DB 174, and a comment DB 175.

The validity estimation model 171 is an estimator that estimates the validity of a comment based on the learning situation data of a learner and a comment made by an instructor on the learning situation data, and is a trained learning model generated by machine learning. The training data DB 172 stores training data for constructing (creating) the validity estimation model 171. The learning model management DB 173 stores information related to the trained validity estimation model 171. The learner DB 174 stores information related to learners. The comment DB 175 stores comments that can be used by instructors.

In this embodiment, the memory unit 12 and the large-capacity memory unit 17 may be configured as an integrated memory device. The large-capacity memory unit 17 may also be configured from multiple memory devices. Furthermore, the large-capacity memory unit 17 may be an external memory device connected to the server 1.

Server 1 may be executed on a single computer, or may be executed in a distributed manner on multiple computers, or may be executed in a distributed manner on virtual machines.

Figure 3 is an explanatory diagram showing an example of a record layout of training data DB172. Training data DB172 includes a training ID column, an input column, an output column, and a personality type column.

The training ID column stores a unique ID of the training data to identify each training data. The input column includes a learning status column and a comment column. The learning status column stores time-series learning status data indicating the learning status of the learner. The learning status data is a numerical value quantified by the learner's learning amount, academic ability deviation value, score, or a combination of these. The learning amount is, for example, the number of learning tasks within a specified period (e.g., one month), or the learning time within a specified period. For example, when using a learning amount that combines the number of tasks and the learning time, the learning amount for 10 questions is set to one hour of learning time, and the learning status data becomes 10 (10 x 1). The learning status data may be quantified based on any calculation formula.

The comment column stores the instructor's comments on the learning situation data. The output column stores information indicating the validity of the comments. The personality type column stores the learner's personality type. Personality types will be described later.

FIG. 4 is an explanatory diagram showing an example of the record layout of the learning model management DB 173 and the learner DB 174.
The learning model management DB173 includes a model ID column, a learning model column, a type column, and a generation date and time column. The model ID column stores a uniquely specified ID of the learning model to identify each trained learning model. The learning model column stores a file of the trained learning model. The type column stores the type of learning model. The type column stores, for example, the type of learning model generated based on the learner's personality type. For example, the type column stores "personality type A model" or "personality type B model", etc. The generation date and time column stores date and time information when the learning model was generated.

The learner DB174 includes a learner ID column and a personality type column. The learner ID column stores a unique learner ID to identify each learner. The personality type column stores a personality type indicating the learner's personality. With regard to the personality type, for example, a personality type diagnosis is performed on each learner to obtain the personality type of each learner, and the obtained personality type is associated with the learner ID and registered in the learner DB174.

In regard to the diagnosis of the personality type of the learners, for example, a personality diagnosis method called Ecogram may be used to calculate scores for personality traits such as "strictness," "kindness," "calmness," "free-spiritedness," and "obedience," and based on the calculated scores, it may be diagnosed which personality trait best applies to each learner. Alternatively, a personality diagnosis method called OK-gram may be used to calculate scores for personality traits such as "negativity toward others," "positive toward others," "negativity toward self," and "positive toward self," and based on the calculated scores, it may be diagnosed which personality trait best applies to each learner. Furthermore, for example, several questions may be asked of each learner in advance, and information obtained based on the answers to the questions may be analyzed to diagnose which personality trait each learner has. For example, personality types include "type that grows when praised" and "type that grows when strict," etc.

Figure 5 is an explanatory diagram showing an example of a record layout of comment DB175. Comment DB175 includes a comment ID column and a comment column. The comment ID column stores a unique comment ID to identify each comment. The comment column stores the content of the instructor's comment.

Note that the storage format of each DB described above is just an example, and other storage formats are also possible as long as the relationships between data are maintained.

Figure 6 is a block diagram showing an example of the configuration of the learner terminal 2. The learner terminal 2 includes a control unit 21, a memory unit 22, a communication unit 23, an input unit 24, and a display unit 25. Each component is connected by a bus B.

The control unit 21 includes an arithmetic processing device such as a CPU or MPU, and performs various information processing, control processing, etc. related to the learner terminal 2 by reading and executing the control program 2P stored in the memory unit 22. Note that while FIG. 6 describes the control unit 21 as a single processor, it may also be a multi-processor. The memory unit 22 includes memory elements such as RAM and ROM, and stores the control program 2P or data, etc. required for the control unit 21 to execute processing. The memory unit 22 also temporarily stores data, etc. required for the control unit 21 to execute arithmetic processing.

The communication unit 23 is a communication module for performing communication-related processing, and transmits and receives information to and from the server 1 via the network N. The input unit 24 may be a keyboard, a mouse, or a touch panel integrated with the display unit 25. The display unit 25 is a liquid crystal display or an organic EL display, etc., and displays various information according to instructions from the control unit 21.

Figure 7 is a block diagram showing an example of the configuration of the instructor terminal 3. The instructor terminal 3 includes a control unit 31, a storage unit 32, a communication unit 33, an input unit 34, and a display unit 35. Each component is connected by a bus B.

The control unit 31 includes an arithmetic processing device such as a CPU or MPU, and performs various information processing, control processing, etc. related to the instructor terminal 3 by reading and executing the control program 3P stored in the memory unit 32. Note that while FIG. 7 describes the control unit 31 as a single processor, it may also be a multi-processor. The memory unit 32 includes memory elements such as RAM and ROM, and stores the control program 3P or data, etc. required for the control unit 31 to execute processing. The memory unit 32 also temporarily stores data, etc. required for the control unit 31 to execute arithmetic processing.

The communication unit 33 is a communication module for performing communication-related processing, and transmits and receives information to and from the server 1 via the network N. The input unit 34 may be a keyboard, a mouse, or a touch panel integrated with the display unit 35. The display unit 35 is a liquid crystal display or an organic EL display, etc., and displays various information according to instructions from the control unit 31.

Next, we will explain the process of estimating the validity of the instructor's comments based on the learner's time-series learning status data and the instructor's comments on the learning status data. In this embodiment, the validity estimation model 171 is used to estimate the validity of the comments, and the estimation result is output.

Figure 8 is an explanatory diagram of the validity estimation model 171. In this embodiment, the server 1 performs machine learning to learn the relationship between the learner's time-series learning situation data, the instructor's comments on the learning situation data, and the validity of the comments, and generates (creates) the validity estimation model 171.

For example, the validity estimation model 171 is a neural network generated by deep learning, and is an attention model, which is a type of recurrent neural network (RNN).

The attention model-related validity estimation model 171 is classified into a network that functions as an encoder and a network that functions as a decoder, as shown in FIG. 8. The label <EOS (End of String)> is a reserved word that indicates the end of data during learning or estimation processing.

The Encoder is composed of multiple neurons called LSTM (Long-Short Term Memory) blocks. The Encoder accepts input of learning status data for a specified period (e.g., one month or one week) and comments on the learning status data, and performs calculations in the LSTM block. The LSTM block performs calculations while referring to the calculation results in the previous LSTM block (neuron), i.e., the calculation results based on the learning status data and comments that appear immediately before. As a result, the validity estimation model 171 estimates the validity of the comments corresponding to the learning status data based on the chronological order.

The instructor's comments may be single or multiple. An example of a single comment is "Try harder." An example of multiple comments is "Good job" and "You worked really hard." The Encoder passes the hidden states associated with the input data at each time to the Attention layer.

The Attention Layer receives the hidden state vector for each time passed from the Encoder. The Attention Layer calculates the score of the hidden state vector for each time input from the Encoder. For example, the score may be calculated using a softmax function. By multiplying each hidden state vector by the score passed through softmax, it is possible to amplify hidden states with high scores and eliminate hidden states with low scores.

The Attention layer passes the converted hidden state vector to the Decoder. The Decoder is composed of neurons called LSTM blocks. The Decoder adds up the hidden state vectors for each time input from the Attention layer, combines them with the Decoder's hidden state vector, converts them, and obtains the Decoder's final hidden state vector. The Decoder outputs an estimation result that estimates the validity of the comment based on the final hidden state vector. The estimation result output from the Decoder is, for example, a continuous probability value indicating validity (for example, a value in the range from "0" to "1").

As shown in the figure, the Encoder accepts input of time-series learning status data (e.g., learning status data for April, May, and June) and comments on the learning status data (e.g., try harder), and performs calculations in the middle layer. Specifically, the server 1 divides the comment into multiple segments. A segment is a component of a sentence, and is the smallest unit obtained when dividing a sentence to an extent that does not sound unnatural as actual words. For example, if the comment is "try harder," the server 1 divides "try harder" into three segments including "more," "try harder," and "ne." The server 1 combines each segment with the time-series learning status data. For example, the "more" segment is combined with the learning status data for April, the "try harder" segment is combined with the learning status data for May, and the "ne" segment is combined with the learning status data for June. The server 1 sequentially inputs multiple data combining segments and learning status data into each LSTM block. In FIG. 8, monthly learning status data is described, but this is not limited to this, and learning status data may be, for example, weekly learning status data.

The Encoder passes the hidden state vector associated with the input data for each time to the Attention layer. The Attention layer converts the hidden state vector for each time and passes the converted hidden state vector to the Decoder. The Decoder combines and converts the hidden state vector for each time input from the Attention layer with the Decoder's hidden state vector, and outputs an estimation result that estimates the validity of the comment.

The attention model is an example of the validity estimation model 171, and for example, the server 1 may generate other neural networks such as Seq2Seq (Sequence to Sequence) as the validity estimation model 171. In addition, the validity estimation model 171 is not limited to a neural network, and may be a model based on other learning algorithms, such as a decision tree, a random forest, or an SVM (Support Vector Machine).

For example, the server 1 generates the validity estimation model 171 using training data stored in the training data DB 172. The training data is a combination of time-series learning status data of a learner and comments on the learning status data, which are associated with information indicating the validity of the comments. The training data is generated based on a large amount of learning status data collected from learners and comments from instructors corresponding to the learning status data. The training data may be data manually created separately.

The server 1 inputs the time-series learning situation data for training and the instructor's comments as input data to the validity estimation model 171, and obtains the estimation result (output data) of the validity of the comments from the output layer. The server 1 compares the correct output data contained in the training data with the output data output from the validity estimation model 171, and optimizes various parameters such as weights used in calculations in the intermediate layer so that the two are close to each other.

As a result, the server 1 generates (constructs) the validity estimation model 171. The server 1 stores the generated validity estimation model 171 in the learning model management DB 173. In this embodiment, the server 1 uses the validity estimation model 171 to estimate the validity of the comment on the learning situation data of the learner, and outputs the estimation result of the validity of the comment on the learning situation data and the instructor's comment.

As shown in the figure, for a comment corresponding to the learning status data, an inference result is output in which the probability value of "appropriate" is "0.90" and the probability value of "inappropriate" is "0.10". In addition, an inference result may be output using a predetermined threshold. For example, when the server 1 determines that the probability value of "appropriate" (0.90) is equal to or greater than a predetermined threshold (e.g., 0.85), it outputs an inference result that the comment is appropriate. Alternatively, when the server 1 determines that the probability value of "inappropriate" (0.87) is equal to or greater than a predetermined threshold (e.g., 0.85), it outputs an inference result that the comment is inappropriate.

Furthermore, when the instructor's comments include multiple comments, the server 1 uses the validity estimation model 171 to output the estimation result of estimating the validity of each comment. For example, when the instructor's comments include "Good job" and "You worked really hard", the server 1 uses the validity estimation model 171 to output both the estimation result of the validity of "Good job" and the estimation result of the validity of "You worked really hard".

For example, for the comment "Good job," server 1 outputs an inference result in which the probability value of "appropriate" is "0.90" and the probability value of "inappropriate" is "0.10." In addition, for the comment "You worked really hard," server 1 outputs an inference result in which the probability value of "appropriate" is "0.15" and the probability value of "inappropriate" is "0.85." In this case, server 1 outputs the inappropriate comment "You worked really hard" as a comment to be corrected based on the inference result of the instructor's comment.

Figure 9 is a flowchart showing the steps of the process for generating the validity estimation model 171. The process for generating the validity estimation model 171 by performing machine learning will be described with reference to Figure 9.

The control unit 11 of the server 1 acquires multiple pieces of training data from the training data DB 172 of the mass storage unit 17, in which the learner's time-series learning status data and comments on the learning status data are associated with information indicating the validity of the comments (step S101). Based on the acquired time-series learning status data and comments on the learning status data, and information indicating the validity of the comments, the control unit 11 generates a validity estimation model 171 that takes the time-series learning status data and the learning status data as inputs and outputs an estimation result that estimates the validity of the comments (step S102).

For example, the validity estimation model 171 is a neural network related to an attention model, which receives learning status data and comments on the learning status data as input and outputs an estimation result estimating the validity of the comments. The control unit 11 inputs the time-series learning status data and comments to the validity estimation model 171, and outputs an estimation result estimating the validity of the comments. The control unit 11 compares the output validity estimation result with correct answer information indicating validity contained in the training data, and optimizes various parameters of the intermediate layer so that the two are close to each other, thereby generating the validity estimation model 171.

The control unit 11 stores the generated validity estimation model 171 in the learning model management DB 173 of the mass storage unit 17 (step S103), and ends the series of processes. Specifically, the control unit 11 assigns a model ID to the generated validity estimation model 171, and stores the file and generation date and time of the validity estimation model 171 as a single record in the learning model management DB 173 in association with the assigned model ID.

In addition, a validity estimation model 171 can be generated for each preset personality type that indicates the personality of the learner. Specifically, the server 1 acquires training data for each personality type of the learner from the training data DB 172. For example, when generating a validity estimation model 171 for personality type A, the server 1 extracts multiple pieces of training data corresponding to the personality type that is "personality type A" from the training data DB 172. The server 1 generates a validity estimation model 171 for personality type A using the extracted training data.

Next, we will explain how to use the generated validity estimation model 171.

The learner terminal 2 accepts input of time-series learning status data (e.g., the learner's score, academic ability deviation value, or amount of learning) indicating the learner's learning status. For example, the learner terminal 2 accepts the learner's learning amount for three months (e.g., April, May, and June). If the learner uses an e-learning system that provides assignments, for example, the learner terminal 2 may acquire the learning status data of the learner stored in the e-learning system. The learner terminal 2 transmits the acquired learning status data of the learner to the server 1. The server 1 receives the time-series learning status data of the learner transmitted from the learner terminal 2, and transmits the received learning status data to the instructor terminal 3. The learning status data may be transmitted directly from the learner terminal 2 to the instructor terminal 3.

The instructor terminal 3 receives the learning status data of the learner transmitted from the server 1, and acquires the instructor's comments corresponding to the received learning status data. For example, the instructor terminal 3 acquires multiple pre-registered comments from the comment DB 175, and displays the acquired multiple comments on the screen. The instructor terminal 3 may accept a comment selection from the multiple comments displayed on the screen. The instructor terminal 3 may also directly accept input of the instructor's comments. The instructor terminal 3 transmits the acquired instructor's comments to the server 1 together with the learning status data.

The server 1 estimates the validity of the comment based on the learning status data of the learner transmitted from the instructor terminal 3 and the comment corresponding to the learning status data. Specifically, the server 1 inputs the time-series learning status data and the comment corresponding to the learning status data to the validity estimation model 171, and outputs the estimation result of the validity of the comment.

If the server 1 determines that the comment is valid based on the output estimation result, it transmits the comment estimation result to the instructor terminal 3. The instructor terminal 3 receives the estimation result of the validity of the comment transmitted from the server 1, and displays the received estimation result on the screen. The server 1 transmits the comment to the learner terminal 2. The learner terminal 2 displays the comment transmitted from the server 1 on the screen together with the learning status data.

The validity of multiple comments on the learning status data can be estimated using the validity estimation model 171. Specifically, the server 1 inputs the learning status data and multiple comments on the learning status data into the validity estimation model 171, and outputs an estimation result estimating the validity of each comment. Based on the output validity estimation result, the server 1 acquires inappropriate comments as comments to be corrected. The server 1 transmits the validity estimation result and the comments to be corrected to the instructor terminal 3. The instructor terminal 3 receives the validity estimation result and the comments to be corrected transmitted from the server 1, and displays the received estimation result and the comments to be corrected on the screen.

Figure 10 is an explanatory diagram showing an example of a display screen for the estimated result of the validity of a comment. The screen includes a learning status data display field 11a, a comment selection combo box 11b, a selection button 11c, a comment display field 11d, a check button 11e, and a check result display field 11f.

The learning status data display field 11a is a display field that displays the learning status data of the learner. The comment selection combo box 11b is a combo box that accepts the selection of the instructor's comment on the learning status data. Note that this is not limited to a combo box, and may be, for example, a text field that accepts manual input of the instructor's comment. The selection button 11c is a button that accepts the selection of the instructor's comment. The comment display field 11d is a display field that displays the selected comment. The check button 11e is a button that checks the validity of the comment. The check result display field 11f is a display field that displays the results of checking the validity of the comment.

The instructor terminal 3 acquires the learner's time-series learning status data from the server 1. The instructor terminal 3 generates a graph showing the time-series learning status data based on the acquired learning status data. Note that the graph showing the time-series learning status data may be generated on the server 1 side. The instructor terminal 3 displays the generated graph in the learning status data display field 11a.

As shown, a vertical bar graph showing the relationship between learning status data and time period is displayed in the learning status data display area 11a. The horizontal axis of the vertical bar graph indicates the time period (e.g., April to February of the following year). The vertical axis of the vertical bar graph indicates the learning status data (e.g., amount of learning or time of learning). Note that while a vertical bar graph is used in FIG. 10, this is not limiting and other graph formats may be used.

The instructor terminal 3 acquires multiple pre-set comments for the instructor from the comment DB 175 of the server 1, and sets the acquired multiple comments in the comment selection combo box 11b. For example, the instructor terminal 3 acquires "Try harder," "Nice job," and "You worked really hard," which are pre-registered in the comment DB 175, and sets them in the comment selection combo box 11b. When the instructor terminal 3 receives a touch (click) operation of the selection button 11c, it acquires the comment selected by the comment selection combo box 11b. The instructor terminal 3 displays the comment of the selected instructor in the comment display field 11d.

It is also possible to select multiple comments. For example, when the instructor terminal 3 receives the selection of "Good job" from the comment selection combo box 11b, it displays the received "Good job" in the comment display field 11d. When the instructor terminal 3 receives the selection of "You worked really hard" from the comment selection combo box 11b, it displays the received "You worked really hard" in the comment display field 11d on the line following "Good job."

When the instructor terminal 3 receives a touch operation of the check button 11e, it transmits the selected comment to the server 1 together with the learning status data of the learner. The server 1 estimates (checks) the validity of the comment using the validity estimation model 171 based on the learning status data transmitted from the instructor terminal 3 and the comment corresponding to the learning status data. The server 1 transmits the estimation result (check result) of the validity of the comment to the instructor terminal 3.

The instructor terminal 3 receives the inference results of the comments sent from the server 1, and displays the received inference results in the check result display area 11f. As shown in the figure, for the selected comments "Good job" and "You worked really hard," the inference results that "Good job" is appropriate and "You worked really hard" is inappropriate (unreasonable) are displayed in the check result display area 11f. Inappropriate comments may be displayed in bold or italics, for example. Furthermore, for inappropriate comments, the instructor terminal 3 may output a message indicating that the comment is inappropriate (for example, "This advice is not suitable for learner A") in the check result display area 11f.

Figures 11 and 12 are flowcharts showing the processing steps when performing the process of estimating the validity of a comment. The control unit 21 of the learner's terminal 2 acquires time-series learning status data indicating the learning status of the learner (step S211). For example, the control unit 21 may accept input of the time-series learning status data via the input unit 24. Alternatively, when the learner uses an e-learning system, the control unit 21 may acquire the learning status data of the learner stored in the e-learning system via the communication unit 23. The control unit 21 transmits the learner ID and the acquired learning status data of the learner to the server 1 via the communication unit 23 (step S212).

The control unit 11 of the server 1 receives the learner ID and learning status data transmitted from the learner terminal 2 via the communication unit 13 (step S111), and transmits the received learner ID and learning status data to the instructor terminal 3 (step S112). The control unit 31 of the instructor terminal 3 receives the learner ID and learning status data transmitted from the server 1 via the communication unit 33 (step S311), and displays the received learner ID and learning status data on the display unit 35 (step S312).

The control unit 31 acquires a number of pre-set comments for the instructor from the comment DB 175 of the mass storage unit 17 (step S313). The control unit 31 displays the acquired number of comments on the display unit 35 (step S314). The control unit 31 may directly accept input of the instructor's comments through the input unit 34. The control unit 31 accepts a selection of a comment corresponding to the learning status data from the displayed number of comments through the input unit 34 (step S315). The control unit 31 transmits the learner ID, learning status data, and the selected comment to the server 1 through the communication unit 33 (step S316).

The control unit 11 of the server 1 receives the learner ID, learning status data, and comments transmitted from the instructor terminal 3 via the communication unit 13 (step S113). Based on the received learner ID, the control unit 11 acquires the learner's personality type from the learner DB 174 of the mass storage unit 17 (step S114). Based on the acquired personality type, the control unit 11 identifies the validity estimation model 171 (step S115). Specifically, based on the acquired personality type, the control unit 11 acquires and identifies the relevant validity estimation model 171 from the learning model management DB 173 of the mass storage unit 17.

Based on the received learning status data and comment, the control unit 11 estimates the validity of the comment using the identified validity estimation model 171 (step S116). Specifically, the control unit 11 inputs the learning status data and the comment to the validity estimation model 171 and outputs an estimation result estimating the validity of the comment. Note that when the control unit 11 inputs multiple comments on the learning status data to the validity estimation model 171, it outputs an estimation result estimating the validity of each comment.

The control unit 11 judges whether the comment is appropriate or not based on the output estimation result (step S117). Note that if the instructor's comments include multiple comments, the control unit 11 judges whether each comment is appropriate or not. If the control unit 11 judges that any of the comments is inappropriate, it judges that the instructor's comments are inappropriate.

If the control unit 11 determines that the comment is inappropriate (NO in step S117), it transmits the comment estimation result and the comment to be corrected (the inappropriate comment) to the instructor terminal 3 via the communication unit 13 (step S118). The control unit 31 of the instructor terminal 3 receives the estimation result and the comment to be corrected transmitted from the server 1 via the communication unit 33 (step S317), and displays the received estimation result and the comment to be corrected on the display unit 35 (step S318).

If the control unit 11 determines that the comment is valid (YES in step S117), it transmits the comment estimation result to the instructor terminal 3 via the communication unit 13 (step S119), and transmits the comment to the learner terminal 2 (step S120).

The control unit 31 of the instructor terminal 3 receives the estimation result transmitted from the server 1 via the communication unit 33 (step S319) and displays the received estimation result on the display unit 35 (step S320). The control unit 21 of the learner terminal 2 receives the comment transmitted from the server 1 via the communication unit 23 (step S213). The control unit 21 displays the learning status data and the received comment on the display unit 25 (step S214) and ends the process.

According to this embodiment, it is possible to estimate the validity of an instructor's comments corresponding to a learner's learning situation data using the validity estimation model 171.

In this embodiment, by creating a validity estimation model 171 for each personality type of a learner, the validity of a comment is estimated for each personality type, making it possible to generate appropriate comments for the learner.

(Embodiment 2)
The second embodiment relates to a form in which an instructor's comments are generated using artificial intelligence based on the learning situation data of the learner. Note that a description of the contents that overlap with the first embodiment will be omitted.

Figure 13 is a block diagram showing an example of the configuration of the server 1 of embodiment 2. Contents that overlap with those in Figure 2 are given the same reference numerals and will not be described. The mass storage unit 17 stores a comment estimation model (second learning model) 176. The comment estimation model 176 is an estimator that estimates the instructor's comments based on the learning situation data of the learner, and is a trained learning model generated by machine learning. The training data DB 172 stores training data for constructing the comment estimation model 176. The learning model management DB 173 stores information related to the trained comment estimation model 176.

Figure 14 is an explanatory diagram showing an example of a record layout of the training data DB 172 in embodiment 2. Note that explanations of contents that overlap with FIG. 3 will be omitted. The training data DB 172 includes a training ID column, an input column, an output column, and a personality type column. The input column stores the time-series learning status data of the learner. Note that the learning status data is the same as in embodiment 1, so explanations will be omitted. The output column stores the instructor's comments on the learning status data.

FIG. 15 is an explanatory diagram of the comment estimation model 176. In this embodiment, the server 1 performs machine learning to learn the relationship between the learner's time-series learning status data and the instructor's comments corresponding to the learning status data, and generates (creates) the comment estimation model 176.

For example, the comment prediction model 176 is a neural network generated by deep learning, and is an attention model, which is a type of RNN.

As shown in FIG. 8, the comment estimation model 176 related to the attention model is classified into a network that functions as an encoder and a network that functions as a decoder. The label <EOS> is a reserved word that indicates the end of data during learning or estimation processing.

The Encoder is composed of multiple neurons called LSTM blocks. The Encoder accepts input of learning situation data in a time series and performs calculations in the LSTM blocks. The LSTM blocks perform calculations while referring to the calculation results in the previous LSTM block, i.e., the calculation results based on the learning situation data that appears immediately before. This allows the comment estimation model 176 to estimate the comment that corresponds to the learning situation data based on the time series order.

The Encoder passes the hidden state vectors (hidden states) associated with the input data at each time to the Attention layer. The Attention layer receives the hidden state vectors at each time passed from the Encoder and performs conversion processing on the received hidden state vectors at each time. Note that the conversion processing of the hidden state vectors is the same as that shown in FIG. 8, so a description thereof will be omitted.

The Attention layer passes the converted hidden state vector to the Decoder. The Decoder is composed of neurons called LSTM blocks. The Decoder adds up the hidden state vectors for each time input from the Attention layer, combines them with the Decoder's hidden state vector, converts them, and obtains the Decoder's final hidden state vector. The Decoder outputs an estimation result that estimates the comment corresponding to the learning situation data based on the final hidden state vector.

As shown in the figure, the Encoder accepts input of time-series learning status data (e.g., learning status data for April, May, and June) and performs calculations in the intermediate layer. The Encoder passes the hidden state vector associated with the input data for each time to the Attention layer. The Attention layer converts the hidden state vector for each time and passes the converted hidden state vector to the Decoder. The Decoder combines and converts the hidden state vector for each time input from the Attention layer with the Decoder's hidden state vector, and outputs a comment corresponding to the learning status data (e.g., "Try harder.").

Note that the attention model is one example of the comment estimation model 176, and for example, the server 1 may generate other neural networks such as Seq2Seq as the comment estimation model 176. In addition, the comment estimation model 176 is not limited to a neural network, and may be a model based on other learning algorithms, such as a decision tree, a random forest, or an SVM.

For example, the server 1 generates a comment estimation model 176 using training data stored in the training data DB 172. The training data is a combination of time-series learning status data of a learner and comments corresponding to the learning status data. The training data is generated based on a large amount of learning status data collected from learners and comments by instructors corresponding to the learning status data. The training data may be data manually created separately.

The server 1 inputs the time-series learning status data for training as input data to the comment estimation model 176, and obtains the estimation result (output data) of the comment corresponding to the learning status data from the output layer. The server 1 compares the correct output data included in the training data with the output data output from the comment estimation model 176, and optimizes various parameters such as weights used for calculations in the intermediate layer so that the two are close to each other.

As a result, the server 1 generates a comment estimation model 176. The server 1 stores the generated comment estimation model 176 in the learning model management DB 173. In this embodiment, the server 1 uses the comment estimation model 176 to estimate comments for the learning status data of the learner, and outputs an estimation result of a comment corresponding to the learning status data.

In addition, a comment estimation model 176 can be generated for each personality type of a learner. Specifically, the server 1 acquires training data for each personality type of a learner from the training data DB 172. For example, when generating a comment estimation model 176 for personality type A, the server 1 extracts multiple pieces of training data corresponding to the personality type that is "personality type A" from the training data DB 172. The server 1 generates a comment estimation model 176 for personality type A using the extracted training data.

Figure 16 is an explanatory diagram showing an example of a display screen for the comment estimation results. The screen includes a learning status data display field 12a, a generate button 12b, and a comment display field 12c. The learning status data display field 12a is a display field that displays the learning status data of the learner. The generate button 12b is a button that generates (estimates) a comment. The comment display field 12c is a display field that displays the generated comment.

The instructor terminal 3 acquires the learner's time-series learning status data from the server 1. The instructor terminal 3 generates a graph showing the time-series learning status data based on the acquired learning status data. Note that the graph showing the time-series learning status data may be generated on the server 1 side. The instructor terminal 3 displays the generated graph in the learning status data display field 12a. Note that the display of the graph is the same as in FIG. 10, so a description thereof will be omitted.

When the instructor terminal 3 receives a touch operation of the generate button 12b, it transmits the time-series learning status data to the server 1. The server 1 inputs the time-series learning status data transmitted from the instructor terminal 3 into the comment estimation model 176 and outputs the instructor's comments corresponding to the learning status data. The server 1 transmits the output instructor's comments to the instructor terminal 3. The instructor terminal 3 receives the instructor's comments transmitted from the server 1 and displays them in the comment display field 12c.

Figure 17 is a flowchart showing the processing steps when performing comment generation processing. The control unit 11 of the server 1 acquires the learner's time-series learning status data (step S131). For example, the control unit 11 may acquire the learner's learning status data from the learner's terminal 2 via the communication unit 13. Alternatively, when the learner uses an e-learning system, the control unit 11 may acquire the learner's learning status data stored in the e-learning system.

The control unit 11 transmits the learner ID and the acquired time-series learning status data to the instructor terminal 3 via the communication unit 13 (step S132). The control unit 31 of the instructor terminal 3 receives the learner ID and learning status data transmitted from the server 1 via the communication unit 33 (step S331). The control unit 31 displays the received learner ID and learning status data on the display unit 35 (step S332).

When the control unit 31 receives a comment generation operation corresponding to the received learning status data through the input unit 34 (step S333), the control unit 31 transmits a comment generation request to the server 1 through the communication unit 33 (step S334). For example, when the control unit 31 receives a touch operation of a button for generating a comment through the input unit 34, the control unit 31 generates a comment generation request. The generation request includes the learner ID and the time-series learning status data. The control unit 31 transmits the generated comment generation request to the server 1 through the communication unit 33.

The control unit 11 of the server 1 receives the comment generation request sent from the instructor terminal 3 via the communication unit 13 (step S133). The control unit 11 acquires the learner's personality type from the learner DB 174 of the mass storage unit 17 based on the learner ID included in the received generation request (step S134). The control unit 11 identifies a comment estimation model 176 based on the acquired personality type (step S135). Specifically, the control unit 11 acquires and identifies the relevant comment estimation model 176 from the learning model management DB 173 of the mass storage unit 17 based on the acquired personality type.

The control unit 11 inputs the time-series learning situation data included in the received generation request into the identified comment estimation model 176, and generates (estimates) the instructor's comments corresponding to the learning situation data (step S136). The control unit 11 transmits the generated instructor's comments to the instructor terminal 3 via the communication unit 13 (step S137). The control unit 31 of the instructor terminal 3 receives the instructor's comments transmitted from the server 1 via the communication unit 33 (step S335). The control unit 31 displays the received instructor's comments on the display unit 35 (step S336), and ends the process.

According to this embodiment, it is possible to generate instructor comments corresponding to the learning status data based on the time-series learning status data of the learner using the comment estimation model 176.

According to this embodiment, a comment estimation model 176 is created for each learner's personality type, and the instructor's comments are estimated for each personality type, which is useful for creating appropriate comments.

According to this embodiment, by outputting comments that need to be corrected from among multiple comments on the learning situation data, it becomes possible to help create appropriate comments.

According to this embodiment, by automatically generating the instructor's comments, it is possible to significantly reduce the workload of the instructor.

The embodiments disclosed herein are illustrative in all respects and should not be considered limiting. The scope of the present invention is indicated by the claims, not by the meaning described above, and is intended to include all modifications within the scope and meaning equivalent to the claims.

1. Information processing device (server)
REFERENCE SIGNS LIST 11 Control unit 12 Storage unit 13 Communication unit 14 Input unit 15 Display unit 16 Reading unit 17 Large-capacity storage unit 171 Validity estimation model (first learning model)
172 Training Data DB
173 Learning model management DB
174 Learner DB
175 Comment DB
176 Comment estimation model (second learning model)
1a Portable storage medium 1b Semiconductor memory 1P Control program 2 Information processing terminal (student terminal)
21 Control unit 22 Storage unit 23 Communication unit 24 Input unit 25 Display unit 2P Control program 3 Information processing terminal (instructor terminal)
31 Control unit 32 Memory unit 33 Communication unit 34 Input unit 35 Display unit 3P Control program

Claims (9)

a receiving unit for receiving time-series learning status data indicating the learning status of a learner and a comment from an instructor regarding the learning status data;
an output unit that inputs the learning status data and comments received by the reception unit into a first learning model that has been trained to estimate the validity of the comment when time-series learning status data for a specified period and a comment on the learning status data are input, and outputs an estimation result that estimates the validity of the comment. inputting learning status data and a plurality of comments on the learning status data into the first learning model, and outputting an estimation result of estimating the validity of each comment;
The information processing apparatus according to claim 1 , further comprising: outputting an inappropriate comment as a comment to be corrected based on the outputted estimation result. 3 . The information processing apparatus according to claim 1 , wherein the first learning model is created for each of predetermined personality types indicating the personality of the learner. 4 . The information processing apparatus according to claim 1 , further comprising: a storage unit that stores the identification information of the learner and the personality type of the learner in association with each other. A plurality of comments from which the comment can be selected are registered,
The information processing apparatus according to claim 1 , wherein the accepting unit accepts a selection of a comment from the plurality of comments. An information processing device as described in any one of claims 1 to 5, further comprising a second output unit that inputs the learning status data received by the reception unit into a second learning model that has been trained to output an instructor's comments on the learning status data when time-series learning status data for a specified period is input, and outputs a comment on the learning status data. The information processing device according to claim 6 , wherein the second learning model is created for each of the learners' personality types set in advance. receiving time-series learning situation data indicating the learning situation of the learner and comments from the instructor regarding the learning situation data;
An information processing method in which a computer executes a process of inputting received learning status data and comments into a first learning model that has been trained to estimate the validity of a comment when time-series learning status data for a specified period and a comment on the learning status data are input, and outputting an estimation result that estimates the validity of the comment. receiving time-series learning situation data indicating the learning situation of the learner and comments from the instructor regarding the learning situation data;
A program that causes a computer to execute a process of inputting received learning status data and comments into a first learning model that has been trained to estimate the validity of a comment when time-series learning status data for a specified period and a comment on the learning status data are input, and outputting an estimation result that estimates the validity of the comment.

Images