JP7129727B1 - Specificity detection device, specificity detection method, specificity detection program and specificity detection system - Google Patents

Abstract

Kind Code: A1 A novel technique is provided for detecting a change from normal that appears to a user. A singularity detection device includes a processor and a storage device. The processor acquires, from the user terminal, the history of the user's operation including information representing the position at which the user has operated on the predetermined program or information representing the time from the predetermined reference time to the user's operation, and stores the history. machine-learning the features of the score calculated using the history of operations accumulated in the storage device to create a learned model; Calculates an index value according to the degree of divergence from the score calculated based on the operation of . [Selection drawing] Fig. 2

Description

The present invention relates to a peculiarity detection device, a peculiarity detection method, a peculiarity detection program, and a peculiarity detection system.

Conventionally, there has been proposed an information processing device that highly accurately predicts the attendance risk caused by the stress of employees in an organization (eg, Patent Document 1). In the present technology, the storage unit stores stress check data indicating results of stress checks of a plurality of subjects, attendance data of the subjects, and biological sample data of the subjects by multivariate analysis. Prediction weighting parameters and attendance impact risk prediction weighting parameters are stored. The control unit applies the stress prediction weighting parameter to the stress check data and the attendance data of the employee belonging to the organization to generate stress prediction data that predicts the stress of the employee after a predetermined period from the present. to generate Furthermore, the control unit predicts the attendance impact risk including the risk of attendance, productivity decline, or retirement of the employee after the predetermined period of time by applying the attendance impact risk prediction weighting parameter to the stress prediction data. Generate and output attendance impact risk prediction data.

JP 2020-52757 A

For example, when predicting attendance impact risk, even if parameters generated by multivariate analysis based on subject's biological sample data are applied, the accuracy of prediction based on the results of stress checks and signs appearing in attendance data has limits. SUMMARY OF THE INVENTION Accordingly, an object of the present invention is to provide a novel technique for detecting a change from normal that appears to a user.

The uniqueness detection device includes a processor and a storage device. The processor acquires, from the user terminal, a history of the user's operation including information representing the position of the user's operation on a predetermined program or information representing the time from a predetermined reference time to the user's operation, and stores the history. , a learned model is created by machine-learning the features of the score calculated using the operation history accumulated in the storage device, and the predicted value of the score calculated using the learned model and the user Calculates an index value according to the degree of divergence from the score calculated based on the operation of .

In addition, the position operated by the user is the position on the content displayed on the user terminal by a predetermined program or the position on the screen of the user terminal, and the score is an outlier based on the history of past operations. It may change depending on whether or not it is determined to be.

Also, the predetermined reference time is the point in time when a predetermined program causes the user terminal to display the content, and the score may change according to the time from the predetermined reference time to the user's operation. .

Also, the predetermined program may display a predetermined question on the user terminal, and the user's operation may be an operation of answering the predetermined question.

Also, the score may further change according to the percentage of correct answers to a predetermined question.

Further, the predetermined questions may be test questions or stress checks in E-learning, and when the index value exceeds a predetermined threshold, a notification may be given to a predetermined device.

Further, the score may further change according to the amount of history of user's operations acquired in a predetermined period.

The learned model may be created by performing machine learning of score transition features using a neural network.

It should be noted that the contents described in the means for solving the problems can be combined as much as possible without departing from the problems and technical ideas of the present invention. In addition, the content of the means for solving the problem can be provided as a system including a computer or other device, a computer-executed method, or a computer-executed program. A recording medium holding the program may be provided.

It is possible to provide a novel technique for detecting changes from normal that appear to the user.

FIG. 1 is a diagram showing an example of the entire system. FIG. 2 is a block diagram showing an example of the configuration of the peculiarity detection device and the user terminal. FIG. 3 is a processing flow diagram showing an example of the peculiarity detection processing. FIG. 4 is a diagram showing an example of a question table that stores questions. FIG. 5 is a diagram for explaining an example of questions displayed on the user terminal. FIG. 6 is a diagram showing an example of an operation history accumulated in the specificity detection device. FIG. 7 is a diagram showing an example of an answer history accumulated in the specificity detection device. FIG. 8 is a diagram illustrating an example of a heat map that visualizes locations and frequencies of user operations. FIG. 9 is a diagram for explaining an example of an arithmetic expression; FIG. 10 is a diagram showing an example of a screen displaying scores. FIG. 11 is a schematic diagram showing an example of a convolutional neural network. FIG. 12 is a diagram showing an example of a screen displaying predicted values.

EMBODIMENT OF THE INVENTION Hereinafter, the form for implementing this invention is demonstrated with reference to drawings.

<System configuration>
FIG. 1 is a diagram illustrating an example of an entire system according to an embodiment. The system 1 includes user terminals 3 ( 3 A, 3 B) and a specificity detection device 2 , which are connected via a network (communication network) 4 . A network 4 is a communication network such as the Internet, and computers connected to the network 4 can communicate with each other based on various protocols.

The peculiarity detection device 2 is a server that performs peculiarity detection processing according to the present embodiment. The peculiarity detection device 2 transmits to the user terminal 3 predetermined questions about skill checks, stress checks, etc. in E-learning, and accumulates and analyzes the user's operation history and answers to questions received from the user terminal 3. or Moreover, the peculiarity detection device 2 may be realized by sharing the function with a plurality of devices, or may be a device that executes the same processing in parallel.

The user terminal 3 is a general computer such as a smart phone, a tablet, or a PC (Personal Computer), and is used by a user belonging to an organization such as a company. In addition, the user terminal 3 executes a program (also referred to as software or application) according to the present embodiment to display predetermined questions on the screen, display the user's operation history on the user terminal 3, and display information selected by the user. The answer is sent to the peculiarity detection device 2 . A plurality of user terminals 3 may be connected to the network 4 .

<Device configuration>
FIG. 2 is a block diagram showing an example of the configuration of the peculiarity detection device 2 and the user terminal 3. As shown in FIG.

The peculiarity detection device 2 includes an input/output device 21, a storage device 22, a communication interface (I/F) 23, and a processor 24. These components are connected via a bus, for example. . The input/output device 21 is a liquid crystal display, an LED (Light Emitting Diode), a keyboard, a pointing device, or the like. The storage device 22 includes main storage devices such as RAM (Random Access Memory) and ROM (Read Only Memory), and auxiliary storage devices (secondary storage devices) such as HDD (Hard-disk Drive), SSD (Solid State Drive), and flash memory. equipment). The main memory temporarily stores programs read by the processor and secures a work area for the processor. The auxiliary storage stores programs executed by the processor and data transmitted and received to and from the user terminal 3 . The communication I/F 23 is, for example, a network adapter that performs wired or wireless communication, and performs communication based on a predetermined protocol. The processor 24 is an arithmetic processing device such as a CPU (Central Processing Unit), and performs each process according to the present embodiment by executing a program. The example of FIG. 2 shows functional blocks within the processor 24 . Specifically, the processor 24 functions as a questioning control unit 241 , a feature learning unit 242 , a score calculation unit 243 , and a specificity detection unit 244 . The questioning control unit 241 reads predetermined questions related to skill check, stress check, etc. in E-learning, which are stored in advance in the storage device 22, and transmits them to the user terminal 3 via the communication I / F 21, or the user terminal 3 The user's operation history and answers to questions received via the communication I/F 21 are stored in the storage device 22 . The feature learning unit 242 reads out the user's operation history and answers to questions held in the storage device 22, and uses machine learning to learn the features. Also, the feature learning unit 242 causes the storage device 22 to store the model whose features have been learned. The score calculation unit 243 reads out the user's operation history and answers to questions held in the storage device 22 and the feature data output by the feature learning unit 242, and applies them to a predetermined arithmetic expression to calculate a predetermined score. The predetermined score may be, for example, a value that predicts the user's motivation for work. The peculiarity detection unit 244 reads the user's operation history and the answers to the questions stored in the storage device 22, and the feature data output by the feature learning unit 242, and calculates the future score prediction value based on the past data. do. The predicted value may be calculated based on the characteristics of the transition of scores learned based on past data. Also, the score calculated by the score calculation unit 243 is read, and an index value (abnormality degree) indicating the degree of divergence from the predicted value is calculated. If the degree of abnormality is greater than a predetermined standard, it can be said that the user's condition is unique.

The user terminal 3 includes an input/output device 31, a storage device 32, a communication I/F 33, and a processor 34, and these components are connected via signal lines, for example. The input/output device 31 is, for example, a user interface such as a touch panel, and includes an input device and an output device (eg, display device). The storage device 32 is a main storage device such as RAM and ROM, and an auxiliary storage device such as HDD, SSD, and flash memory. The main memory temporarily stores programs read by the processor and secures a work area for the processor. The auxiliary storage stores programs executed by the processor, information related to user operations, and the like. The communication I/F 33 is, for example, a network adapter that performs wired or wireless communication, and performs communication based on a predetermined protocol. The processor 34 is an arithmetic processing device such as a CPU, and performs each process according to the present embodiment by executing a program. The user terminal 3 also shows functional blocks in the processor 34 . Specifically, the processor 34 functions as a questioning control section 341 , an operation recording section 342 and a result display section 343 . The questioning control unit 341 receives predetermined questions from the specificity detection device 2 via the communication I/F 33 and causes the input/output device 31 to display the questions. The operation recording unit also receives user operations via the input/output device 31 and transmits the history of the operations and the answers selected by the user to the specificity detection device 2 via the communication I/F 33 . The result display unit 343 receives results such as scores calculated from the specificity detection device 2 via the communication I/F 33 , stores them in the storage device 32 , or displays them on the input/output device 31 .

<Singularity detection processing>
FIG. 3 is a processing flow diagram showing an example of the peculiarity detection processing. For example, when the user terminal 3 starts the program according to the present embodiment, the user terminal 3 communicates with the specificity detection device 2 . Also, the questioning control unit 241 of the peculiarity detection device 2 transmits a predetermined question to the user terminal 3 ( FIG. 3 : S1).

FIG. 4 is a diagram showing an example of a question table that stores questions. It is assumed that the question table is stored in advance in the storage device 22 of the specificity detection device 2 . The question table includes "question ID", "question", "option 1", "attribute 1", "option 2", "attribute 2", "option 3", "attribute 3", "option 4", Each attribute of "attribute 4" is included. The “question ID” field stores identification information for uniquely identifying a question. The “question” field stores a question sentence displayed to the user. The field of "choices (1 to 4)" stores the answer choices displayed to the user. The "attribute (1 to 4)" field stores information representing the attribute of the corresponding option. The exemplified options include those in which a correct answer or an incorrect answer is determined, such as a skill check in E-learning, and those in which a degree is indicated, such as a stress check. In the example of FIG. 4, the "attribute" field stores information indicating whether or not the answer is correct, and points indicating the degree of accuracy. In S1 of FIG. 3, for example, a predetermined question is extracted from a record registered in such a question table and sent to the user terminal 3. FIG.

On the other hand, the question setting control unit 341 of the user terminal 3 receives a predetermined question and displays it on the input/output device 31 (FIG. 3: S2). FIG. 5 is a diagram for explaining an example of questions displayed on the user terminal. In FIG. 5, the input/output device 31 displays a question 311, an answer option 312, and a "quit" button 313 for ending the skill check. Also, when one of the radio buttons of the option 312 is selected, a "next" button 314 for determining the answer and proceeding to the next question is displayed. In S2 of FIG. 3, such questions are displayed, for example.

For example, when the user operates the input/output device 31 on the screen of FIG. 5, the operation history is recorded. Also, when the user selects one of the radio buttons of option 312 and taps "Next" button 314, the selected answer is recorded. Then, the user's operation history and responses are transmitted to the specificity detection device 2 ( FIG. 3 : S3). On the other hand, upon receiving the user's operation history and answers, the question setting control unit 241 of the peculiarity detection device 2 stores them in the storage device 22 ( FIG. 3 : S4).

FIG. 6 is a diagram showing an example of an operation history accumulated in the specificity detection device. The table in FIG. 6 includes attributes of "user ID", "question ID", "date and time", "operation", and "coordinates". The “user ID” field stores identification information for uniquely identifying the user who operates the user terminal 3 . The “question ID” field stores identification information for uniquely identifying the question displayed to the user. The “date and time” field stores information indicating the date and time when the user performed an operation. Information indicating the operation performed by the user is stored in the “operation” field. The "coordinates" field stores information representing the position on the screen (on the content) where the operation was performed, for example, by a column of coordinates. In other words, the origin of the coordinates may be one of the four corners of the screen of the user terminal 3, or if the content displayed on the screen is of a scrollable size, it may be one of the four corners of the content. It may be in one place. Also, if the operation is a tap, one coordinate is recorded, and if the operation is a swipe, a plurality of coordinate rows representing the trajectory, the coordinates of the start point and the end point, etc. are recorded.

FIG. 7 is a diagram showing an example of an answer history accumulated in the specificity detection device. The table in FIG. 7 includes attributes of "user ID", "question ID", "date and time", and "answer". The “user ID” field stores identification information for uniquely identifying the user who operates the user terminal 3 . The “question ID” field stores identification information for uniquely identifying the question displayed to the user. The “date and time” field stores information indicating the date and time when the user performed an operation. Information indicating the answer selected by the user is stored in the "answer" field.

Further, the questioning control unit 241 of the peculiarity detection device 2 determines whether to end the questioning ( FIG. 3 : S5). For example, the questioning control unit 241 determines to end when all of a predetermined set of questions has been given. If there is a question that has not yet been asked (S5: NO), the questioning control unit 241 returns to S1 and repeats the processing for the question that has not yet been asked.

On the other hand, if it is determined in S5 to end the questioning (S5: YES), the feature learning unit 242 of the peculiarity detection device 2 learns the features of the user's operation history (FIG. 3: S6). This step may be executed when operation histories equal to or greater than a predetermined threshold are accumulated by repeating the processes of S1 to S4. In this step, a model is created that represents the characteristics of the operation history for each user.

For example, a trained model for detecting outliers (abnormal values, peculiarities) is created for each user and each operation screen using One Class SVM (Support Vector Machine). For each operation screen, for example, each question corresponding to each record in FIG. 4 may be displayed. Further, operation histories may be collected not only for question answer screens, but also for application menu screens and the like, so that operation characteristics may be learned in this step.

FIG. 8 is a diagram illustrating an example of a heat map that visualizes locations and frequencies of user operations. The heat map represents the boundaries based on, for example, the density of coordinates tapped by the user in the operation history for the last predetermined number of times. According to the One Class SVM mentioned above,
In such a heat map, it can be said that it is possible to learn the boundary between a place where the user is likely to tap on a certain screen and a place where the user is unlikely to tap.

Also, the learning target in the operation history may be the position tapped or swiped by the user, the time from the display of a screen such as a question to the start of the operation, or a combination thereof. In this way, it is possible to extract the characteristics of what kind of operation the user performs with which finger of the left or right hand when viewing the same type of screen. In addition, it is possible to extract the characteristics of the standard time from when the user sees the same type of screen until he or she comprehends the content and performs the operation of answering.

Alternatively, a model representing the characteristics of the operation history for each user and each operation screen may be created by a clustering method such as the k-means method. That is, a learned model is created that classifies the characteristics of operations performed by the same user on the same type of screen into one cluster. By doing so, it becomes possible to determine that an operation that is not classified into the relevant cluster has specificity.

Also, the score calculation unit 243 of the specificity detection device 2 calculates a score based on a predetermined arithmetic expression, and transmits the score to the user terminal 3 (FIG. 3: S7). FIG. 9 is a diagram for explaining an example of an arithmetic expression; Although FIG. 9 is shown in a tabular form for convenience of explanation, such information does not have to be held as a database table. In this embodiment, different points are added depending on whether or not a predetermined criterion is satisfied, and a total score is calculated for a plurality of criteria. The table in FIG. 9 includes attributes of “scoring reference ID”, “evaluation data”, “learning data”, “True condition”, and “score (True/False)”. In the field of “point addition criterion ID”, identification information for uniquely identifying the point addition criterion is displayed. Data to be evaluated is displayed in the “evaluation data” field. In the field of "learning data", data to be machine-learned and used for comparison with data to be evaluated is displayed. The "True condition" field displays the conditions that must be met in order to add a predetermined number of points. The "score (True/False)" field displays the points added depending on whether the True condition is met or not. In addition, the score is an example and is not limited to the values in FIG.

The record with the "additional reference ID" of C1 gives 20 points if the position operated by the user on the predetermined screen is determined not to be an outlier based on the learned model that has learned the last predetermined number of operation positions. 10 points are added when it is determined to be an outlier. The operation may target tapping, for example. Note that the most recent predetermined number of operations to be learned may not include the latest operation position. It is assumed that the trained model is created in advance by the above-mentioned One Class SVM, k-means method, or other machine learning techniques.

The record with the "additional point ID" of C2 has a time required for the user to perform an operation after being displayed on a predetermined screen compared to the average value of the time required from display to predetermined operation for the most recent predetermined number of cases. 20 points are added when the error is within a predetermined threshold, and 10 points are added when the error is earlier or later than the threshold. Note that the predetermined operation may be the first tap or swipe, or may be an operation to confirm the answer to the question. Also, the threshold may be expressed as a ratio to the average value, or may be an absolute time. Also, the score may be increased or decreased according to the degree of deviation from the reference value such as the average value.

The record with the “scoring reference ID” of C3 indicates that 20 points are added when the accuracy rate for a predetermined set of questions is equal to or higher than the threshold, and 10 points are added when it is below the threshold. Note that the threshold may not be a fixed value but may be expressed as a percentage of the average value. Also, the score may be increased or decreased according to the degree of deviation from the reference value such as the average value.

The record with the "additional reference ID" of C4 gives 10 points if the error in the number of operation records per day is within a predetermined threshold with respect to the average value of the number of operation records per day in the most recent predetermined period. 5 points are added when there is less or more than the threshold value. Note that the predetermined period is, for example, one week, one month, or the like. Also, the number of operation records per day may be compared with an absolute threshold instead of comparing with learning data. Moreover, the data to be evaluated may not be simply the number of cases per day, but may be a moving average per day for the most recent predetermined period, or the like. Also, the threshold may be expressed as a ratio to the average value, or may be an absolute time. Also, the score may be increased or decreased according to the degree of deviation from the reference value such as the average value.

The record with the “scoring reference ID” of C5 adds 10 points if the number of operation records for a given function in one day has an error within a given threshold with respect to the average value for the most recent predetermined period, and the threshold is It shows that 0 points are added when less or more than . The predetermined function may be, for example, an application related to some kind of business such as groupware, or an application for interacting with AI. Note that the predetermined period is, for example, one week, one month, or the like. Also, the number of operation records per day may be compared with an absolute threshold instead of comparing with learning data. Moreover, the data to be evaluated may not be simply the number of cases per day, but may be a moving average per day for the most recent predetermined period, or the like. Also, the threshold may be expressed as a ratio to the average value, or may be an absolute time. Also, the score may be increased or decreased according to the degree of deviation from the reference value such as the average value.

In S7 of FIG. 3, points are accumulated using at least one of the point addition criteria shown in FIG. 9 and the modified example described above, and a predetermined score is calculated. Moreover, points may be further added based on point addition criteria other than those shown in FIG. For example, the user starts the program according to the present embodiment and learns the characteristics of the time period during which the user answers, and if the degree of divergence from the characteristics of the time period in the past results is large, the score will be low. A point-adding criterion may be adopted. In addition, the questions shown in FIG. 4 may further include category information indicating the type of question, and a score subtotal (breakdown) may be calculated for each category in addition to the total score.

On the other hand, when the result display unit 343 of the user terminal 3 receives the calculated score, it causes the input/output device 31 to output it ( FIG. 3 : S8). In this step, the result as shown in FIG. 10 is displayed on the screen of the user terminal 3, for example. FIG. 10 is a diagram showing an example of a screen displaying scores. In the example of FIG. 10, a total score 315, a radar chart 316 based on subtotals calculated for each question category, and an "OK" button 317 for ending the display are displayed on the input/output device 31. .

In addition, the peculiarity detection unit 244 of the peculiarity detection device 2 calculates the predicted value of the future score based on the past score, and calculates the degree of abnormality indicating the degree of divergence between the actual score and the predicted value. , to the user terminal 3 (FIG. 3: S9). In this step, the peculiarity detection unit 244 or the feature learning unit 242 performs machine learning based on changes in scores recorded in the past, and creates a trained model for calculating predicted values of future scores. Machine learning can be performed using, for example, neural networks.

FIG. 11 is a schematic diagram showing an example of a convolutional neural network (CNN). A plurality of nodes included in the input layer of the CNN receives, for example, a total score calculated in past answer operations in chronological order. In the convolution layer, while moving a window of a predetermined size on the nodes of the input layer, the value of the node included in the window (initial score) and the kernel (filter) of a predetermined size are combined. Calculate the sum of the products and convert to a tensor to create a feature map. In the pooling layer, one numerical value representing the feature is generated based on multiple numerical values contained in the window, and the number of data in the layer is reduced. Note that the size (height) of each layer in FIG. 11 indicates the number of data in each layer. A fully connected layer is a multi-layer perceptron that connects all the way through combinations of nodes contained in two layers. The output layer is a predetermined activation function, such as softmax, which outputs an output value representing, for example, probability. For example, the output layer may contain 101 nodes representing probabilities of 100 scores from 0 to full (e.g. 100), 11 nodes represented by . In this way, the score corresponding to the node that outputs the highest output value (probability) can be used as the predicted value. It is also possible to specify a range of nodes whose output values (probabilities) are equal to or greater than a predetermined threshold, and obtain the predicted maximum and minimum predicted scores based on the scores corresponding to the nodes. Note that FIG. 11 schematically shows the structure of the CNN, and the number of layers and the size of each layer are not limited to the example of FIG. For the trained model generated in this way, by changing the teacher value for the input value, for example, a model that outputs a predicted value for the next two days, a model that outputs a predicted value for the next three days, etc. can be created. . Also, a neural network other than CNN may be employed.

In addition, in S9 of FIG. 3, the specificity detection unit 244 obtains the degree of abnormality based on the degree of divergence between the calculated predicted value and the actual score calculated in S7. That is, the greater the difference between the predicted value and the actual score, the greater the degree of anomaly. Moreover, when the degree of abnormality is equal to or greater than a predetermined threshold, the specificity detection unit 244 may determine that the specificity is detected.

According to the detection of specificity, the person himself/herself or related parties can quickly notice that the user is in a state different from normal. For example, when calculating the above-mentioned degree of anomaly and detecting peculiarities based on skill checks, stress checks, and other work-related program operations in E-learning, early detection of signs of user's mental disorders and job separation . The result of detecting the peculiarity may be transmitted to the user terminal 3, or may be notified to a predetermined device (not shown) such as a terminal of a person concerned. That is, by using a score that changes based on the characteristics of the user operating the terminal and the accuracy rate that evaluates the user's ability, it is possible to detect changes in the user from normal times.

Also, in S9, for example, the predicted value is transmitted to the user terminal 3 as well. On the other hand, the result display unit 343 of the user terminal 3 receives the predicted values and the like from the specificity detection device 2 and displays them on the input/output device 31 (FIG. 3: S10). In this step, the result as shown in FIG. 12 is displayed on the screen of the user terminal 3, for example. FIG. 12 is a diagram showing an example of a screen displaying predicted values. In the example of FIG. 12, a graph 318 representing the highest and lowest predicted user scores for the next week, advice on how to spend the day based on the prediction 319, and an "OK" button 317 to end the display. is displayed on the input/output device 31 . Note that the advice 319 may be the result of analyzing the periodicity of the predicted values, or may be content calling attention to the day when the score is predicted to drop.

<Modification>
In addition criteria shown in FIG. 4, the result of negative/positive determination of the user's answer may be used. Negative-positive judgment is a method of decomposing a sentence into words (parts of speech) by, for example, morphological analysis, and determining the judgment result based on the polarities representing the impressions of words registered in a pre-created polarity dictionary and a predetermined arithmetic expression. . Negative/Positive determination can be performed on an answer selected by the user, or by performing natural language processing on text freely input by the user. That is, it becomes possible to set questions without options.

In addition, in the above-described embodiment, an example of detecting specificity based on the operation of a program related to skill check, stress check, and other work in E-learning was shown, but specificity detection can be performed in other programs. can also For example, in an application for elderly people, the peculiarity of a user's operation may be detected to discover signs of dementia. Also, for example, in a game application, the peculiarities of user operations may be detected and used to predict the user's continuation rate or withdrawal rate, or may be used to evaluate the success or failure of some event. Also, for example, in an online shopping application, the peculiarity of a user's operation may be detected and used to select an advertisement to be distributed to the user.

<Others>
The configuration of the embodiment described above is an example, and can be combined and changed as much as possible without departing from the subject and technical idea of the present invention.

The present invention also includes a computer program for performing the processes described above. Alternatively, the program may be provided as a computer-readable recording medium. A computer reads and executes the recording medium on which the program is recorded, thereby enabling the above processing.

Here, a computer-readable recording medium is a recording medium that stores information such as data and programs by electrical, magnetic, optical, mechanical, or chemical action and can be read by a computer. Examples of such recording media that can be removed from the computer include flexible disks, magneto-optical disks, optical disks, magnetic tapes, memory cards, and the like. Recording media fixed to computers include hard disk drives, ROMs, and the like.

1: Singularity detection system 2: Singularity detection device 21: Input/output device 22: Storage device 24: Processor 241: Questioning control unit 242: Feature learning unit 243: Score calculation unit 244: Singularity detection unit 3 (3A, 3B ): User terminal 31: Input/output device 32: Storage device 34: Processor 341: Questioning control unit 342: Operation recording unit 343: Result display unit 4: Network

Claims (11)

including a processor and a storage device;
The processor
obtaining, from a user terminal, a history of user operations, including coordinates on the content, representing positions at which the user has operated a predetermined program for displaying predetermined content, and storing the history in the storage device;
creating a learned model by machine-learning features of a score that changes based on the position operated by the user, which is calculated using the history of the operations accumulated in the storage device;
A specificity detection device that calculates an index value according to a degree of divergence between a predicted value of the score calculated using the trained model and the score calculated based on a user's operation. The peculiarity detection device according to claim 1, wherein the score changes depending on whether or not the position operated by the user is determined to be an outlier based on the history of past operations. The peculiarity detection device according to claim 1 or 2, wherein the score changes according to the time from when the predetermined program causes the user terminal to display the content until the user performs an operation. The predetermined program causes the user terminal to display a predetermined question,
The peculiarity detection device according to any one of claims 1 to 3, wherein the user's operation is an operation of answering the predetermined question. The peculiarity detection device according to claim 4, wherein the score further changes according to the percentage of correct answers to the predetermined question. The predetermined questions are test questions or stress checks in E-learning,
The peculiarity detection device according to claim 4 or 5, wherein when the index value exceeds a predetermined threshold, a predetermined device is notified. The peculiarity detection device according to any one of claims 1 to 6, wherein the score further changes according to the amount of the user's operation history acquired in a predetermined period. The peculiarity detection device according to any one of claims 1 to 7, wherein the trained model is created by machine-learning the transition characteristics of the score using a neural network. said processor of a computer comprising a processor and a storage device,
obtaining, from a user terminal, a history of user operations, including coordinates on the content, representing positions at which the user has operated a predetermined program for displaying predetermined content, and storing the history in the storage device;
creating a learned model by machine-learning the characteristics of the score that changes based on the position operated by the user, which is calculated using the history of the operation accumulated in the storage device;
A method for detecting specificity, wherein an index value is calculated according to a degree of divergence between a predicted value of the score calculated using the trained model and the score calculated based on a user's operation. to said processor of a computer comprising a processor and a storage device;
Acquiring from the user terminal a user operation history including coordinates on the content representing a position where the user has operated a predetermined program for displaying the predetermined content, and storing the history in the storage device;
create a learned model by machine learning the feature of the score that changes based on the position operated by the user, which is calculated using the history of the operation accumulated in the storage device;
A peculiarity detection program for calculating an index value according to the degree of divergence between the predicted score value calculated using the trained model and the score calculated based on a user's operation. including a plurality of servers and a storage device provided by any of the plurality of servers or other devices,
The plurality of servers are
obtaining, from a user terminal, a history of user operations, including coordinates on the content, representing positions at which the user has operated a predetermined program for displaying predetermined content, and storing the history in the storage device;
creating a learned model by machine-learning features of a score that changes based on the position operated by the user, which is calculated using the history of the operations accumulated in the storage device;
A specificity detection system that calculates an index value according to the degree of divergence between the predicted score value calculated using the trained model and the score calculated based on a user's operation.

Images