Authentication device
The authentication device improves mouse authentication accuracy by constructing application-specific learning models for UI operations, addressing the inconsistency caused by varying application layouts.
Patent Information
- Application Number
- JP2025077305
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2025-05-07
- Publication Date
- 2026-11-19
AI Technical Summary
Mouse authentication accuracy is difficult to stabilize due to varying operation characteristics based on different application UI layouts, as suggested by Fitts' Law, leading to inconsistent learning and authentication performance.
An authentication device that constructs machine learning models for each type of application during UI operation, identifies the application being used, and authenticates based on the corresponding learning model to improve accuracy.
Enhances authentication accuracy by learning and applying application-specific mouse operation characteristics, stabilizing performance across different applications.
Smart Images

Figure 2026181728000001_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to an authentication device.
Background Art
[0002] In recent years, research on biometric authentication based on UI (User Interface) operations such as mouse operations has been actively advanced. Among biometric authentication based on UI operations, for example, mouse authentication enables real-time authentication and does not require input of an ID and a password, and thus has the advantages of biometric authentication that the burden on a user is small and the convenience is high. Furthermore, mouse authentication has the advantage that the introduction cost is extremely low since no additional device is required.
Prior Art Documents
Non-Patent Documents
[0003]
Non-Patent Document 1
Summary of the Invention
Problems to be Solved by the Invention
[0004] However, mouse authentication has a problem that the authentication accuracy is difficult to stabilize. For example, even for the feature amount of a mouse operation of the same person, the feature amount strongly depends on a UI (User Interface) such as the layout of buttons on a screen. For this reason, in mouse authentication, if the application to be operated is different, the operation characteristics are also different, and thus learning cannot be performed well. This can be understood from psychological laws such as Fitts' Law (Figure 19) (Reference Document 1), and it has been suggested that this will be important in the future field of mouse authentication. Reference 1: "Fitts' Law and its Application in UX," [Retrieved March 18, 2025], Internet<https: / / u-site.jp / alertbox / fitts-law>
[0005] The present invention has been made in view of the above, and aims to provide an authentication device that can improve the accuracy of authentication based on UI operation. [Means for solving the problem]
[0006] To solve the above-mentioned problems and achieve the objective, the authentication device according to the present invention is characterized by comprising: a learning unit that constructs a machine learning model for each type of application during UI (User Interface) operation, and causes each learning model to learn feature quantities based on the UI operation characteristics of a first user during the execution of the corresponding application; and an authentication unit that identifies which application the UI operation based on the input UI operation information is for, and authenticates whether or not the UI operation based on the input UI operation information is a UI operation by the first user using the learning model corresponding to the identified application. [Effects of the Invention]
[0007] According to the present invention, authentication accuracy based on UI operation can be improved. [Brief explanation of the drawing]
[0008] [Figure 1] Figure 1 is a diagram illustrating the overview of the processing of the authentication device according to the embodiment. [Figure 2] Figure 2 shows an example of the configuration of an authentication device according to an embodiment. [Figure 3] Figure 3 shows the configuration of the authentication unit shown in Figure 2. [Figure 4] Figure 4 is a diagram illustrating the processing of the identification unit. [Figure 5]Figure 5 is a visualization of an example of a mouse trajectory showing mouse movement, along with discrete feature data (Non-Patent Literature 1). [Figure 6] Figure 6 illustrates the direction of mouse movement. [Figure 7] Figure 7 is a flowchart showing the processing procedure of the learning process according to the embodiment. [Figure 8] Figure 8 is a flowchart showing the processing procedure for the authentication process according to the embodiment. [Figure 9] Figure 9 is a diagram illustrating conventional mouse authentication. [Figure 10] Figure 10 is a diagram illustrating a mouse authentication method according to an embodiment. [Figure 11] Figure 11 is a diagram illustrating a mouse authentication method according to an embodiment. [Figure 12] Figure 12 shows an example of an authentication device configuration according to a modified embodiment. [Figure 13] Figure 13 shows an image of key input. [Figure 14] Figure 14 shows the configuration of the authentication unit shown in Figure 12. [Figure 15] Figure 15 shows an example of keystroke information. [Figure 16] Figure 16 is a flowchart showing the processing procedure of the learning process according to a modified example of the embodiment. [Figure 17] Figure 17 is a flowchart showing the processing procedure for the authentication process according to a modified embodiment. [Figure 18] Figure 18 shows an example of a computer in which an authentication device is realized when a program is executed. [Figure 19] Figure 19 illustrates Fitts's Law. [Modes for carrying out the invention]
[0009] [Embodiment] Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. Note that the present invention is not limited by this embodiment. In the description of the drawings, the same parts are denoted by the same reference numerals.
[0010] The embodiment relates to an authentication technology based on UI (User Interface) operation characteristics for operating a computer. In the embodiment, a mouse authentication technology that performs authentication based on the characteristics of mouse operations among UI operations will be described as an example.
[0011] The authentication device according to the embodiment is mounted on, for example, a terminal device. The authorized user of this terminal device is the first user. The authentication device authenticates whether or not the operator of this terminal device is the first user, based on mouse operation information for the terminal device.
[0012] During learning, the authentication device constructs a learning model, which is a machine learning model, for each type of application at the time of mouse operation, and causes each learning model to learn feature quantities based on the mouse operation characteristics of the first user during execution of the corresponding application.
[0013] During authentication, the authentication device identifies which application the mouse operation based on the input mouse operation information is an operation for. Then, using the learning model corresponding to the identified application, the authentication device authenticates whether or not the UI operation based on the input mouse operation information is a UI operation by the first user.
[0014] Figure 1 is a diagram illustrating the overview of the processing of the authentication device according to the embodiment. During learning, the authentication device identifies applications A and B that the first user was using when operating the mouse, and learns the mouse operation characteristics for each application A and B (for example, trajectories R1-a, R1-b (Figure 1)). Here, in order to identify applications A and B, the authentication device additionally acquires information about the active window being operated by the first user. The active window information is data that associates the type of application in the active window with the time.
[0015] The authentication device extracts feature data (mouse features) from the first user's mouse operations for each application A and B, and trains the respective learning models for applications A and B with the mouse features used when operating applications A and B.
[0016] During authentication, the authentication device obtains information about the active window along with mouse operation information, and identifies whether the application being used is application A or application B. Subsequently, the authentication device uses a learning model corresponding to the identified application type to authenticate, based on mouse features, whether the mouse operation was performed by the first user.
[0017] Thus, the embodiment aims to improve authentication accuracy based on mouse operations by acquiring information on the active window along with the mouse operation information of the first user, enabling the learning of mouse features for each application.
[0018] [Authentication device] An authentication device according to an embodiment will now be described. Figure 2 is a diagram showing an example of the configuration of an authentication device according to an embodiment.
[0019] The authentication device 10 according to this embodiment is an authentication device that is realized by loading a predetermined program into a computer including, for example, ROM (Read Only Memory), RAM (Random Access Memory), CPU (Central Processing Unit), etc., and having the CPU execute the predetermined program. The authentication device 10 also has a communication interface for sending and receiving various information with other devices connected via a network or the like.
[0020] The authentication device 10 is installed, for example, in a terminal device operated by a first user. The terminal device has a first input unit 20 and a second input unit 30. The authentication device 10 is connected to the first input unit 20 and the second input unit 30.
[0021] The first input unit 20 inputs AW information related to the active window (AW) operated by the terminal device operator, among the windows displayed on the terminal device's screen, to the authentication device 10.
[0022] The second input unit 30 is, for example, a mouse. The second input unit 30 inputs mouse operation information from the operator of the terminal device to the authentication device 10.
[0023] The authentication device 10 includes an AW logger 11, a mouse logger 12, and a control unit 13.
[0024] The AW logger 11 generates an AW log based on the AW information input from the first input unit 20. The AW log is data that shows the type of AW application in chronological order.
[0025] The mouse logger 12 generates a mouse log based on mouse operation information input from the second input unit 30. The mouse log is data that associates, for example, the coordinates of the cursor pointed to by the mouse, the start and end of left / right clicks, the start and end of extended button clicks, the start and end of up / down wheel rotations, and the time.
[0026] The control unit 13 controls mouse authentication in the authentication device 10. The control unit 13 includes an authentication unit 14 and a learning unit 15.
[0027] The authentication unit 14 uses the AW log and the mouse log to estimate whether the mouse operation based on the mouse log is the mouse operation of the first user.
[0028] The learning unit 15 constructs machine learning models (the first learning model 145-1 to the fourth learning model 145-4 (described later)) for each type of application during mouse operation, and trains each learning model with features based on the UI operation characteristics of the first user while the corresponding application is running.
[0029] The learning unit 15 uses the mouse operation information from the first user and the logs related to the AW during mouse operation by the first user to train the learning model corresponding to the AW application among the various learning models, to learn features based on the mouse operation characteristics while the AW application is running.
[0030] The authentication unit 14 identifies which application the mouse operation based on the input mouse log belongs to. The authentication unit 14 uses the learning model constructed by the learning unit 15 that corresponds to the type of application identified to authenticate whether the mouse operation based on the input mouse log is a mouse operation by the first user.
[0031] If the authentication unit 14 authenticates that the mouse operation is performed by the first user, it permits the user to operate the terminal device. If the authentication unit 14 detects that the mouse operation is not performed by the user, it outputs an alert and does not permit the operation of the terminal device.
[0032] [Authentication Department] Figure 3 shows the configuration of the authentication unit 14 shown in Figure 2. As shown in Figure 3, the authentication unit 14 includes an identification unit 141, an estimation unit 142 (first estimation unit), and a detection unit 146.
[0033] The estimation unit 142 includes feature calculation units 143-1 to 143-4, a first estimation unit 144-1, a second estimation unit 144-2, a third estimation unit 144-3, and a fourth estimation unit 144-4. Feature calculation unit 143-1 and the first estimation unit 144-1 correspond to application A. Feature calculation unit 143-2 and the second estimation unit 144-2 correspond to application B. Feature calculation unit 143-3 and the third estimation unit 144-3 correspond to application C. Feature calculation unit 143-4 and the fourth estimation unit 144-4 correspond to application D. Applications A to D are, for example, Word®, Excel®, PowerPoint®, Slack®, etc.
[0034] The identification unit 141 takes the mouse log (mouse operation information) and the AW log during mouse operation as input. Based on the AW log during mouse operation input along with the mouse log, the identification unit 141 identifies the type of application of the active window corresponding to the input mouse log.
[0035] Figure 4 is a diagram illustrating the processing of the identification unit 141. The active log L1 is data that associates, for example, the active start time of the AW with the type of application displayed on the AW.
[0036] The identification unit 141 divides the mouse operation trajectory R1 shown in the mouse log into applications A, B, C, and D based on the active log L1, according to the start time of AW activity. In other words, the identification unit 141 divides the mouse trajectory R1, which is originally a single stroke, into applications A, B, C, and D of AW (Figure 4 (1)). The identification unit 141 uses logs L11 to L13 of the active log L1 to divide the mouse trajectory R1 into trajectory R11 corresponding to application A (log L11), trajectory R12 corresponding to application B (log L12), and trajectory R13 corresponding to application C (log L13).
[0037] The identification unit 141 separates each mouse log corresponding to the segmented portion of the trajectory R1 according to the type of AW and outputs them to the corresponding feature calculation units 143-1 to 143-4.
[0038] Specifically, the identification unit 141 outputs the mouse log corresponding to trajectory R11 to the feature calculation unit 143-1 corresponding to application A. The identification unit 141 outputs the mouse log corresponding to trajectory R12 to the feature calculation unit 143-2 corresponding to application B. The identification unit 141 outputs the mouse log corresponding to trajectory R13 to the feature calculation unit 143-3 corresponding to application C.
[0039] The estimation unit 142 uses a learning model corresponding to the application identified by the identification unit 141 to estimate whether the mouse operation based on the input mouse log was performed by the first user.
[0040] The feature calculation units 143-1 to 143-4 calculate features (mouse features) based on the input mouse log. For example, the feature calculation units 143-1 to 143-4 calculate features based on the coordinates of the mouse cursor, the start and end of left / right clicks, the start and end of extended button clicks, the start and end of up / down wheel rotations, and the time, all of which are recorded in the mouse log.
[0041] The feature is, for example, the curvature of the mouse's trajectory. Figure 5 is a visualization of an example of a mouse trajectory showing mouse movement, along with discrete feature data (Non-Patent Literature 1). As shown in Figure 5, if the curvature is approximated as a small circular arc, the radius of the curvature changes as it moves along the trajectory based on different data points. Discretely expressed, the curvature K is given by equation (1).
number
[0042] The feature extraction units 143-1 to 143-4 measure specific data points along the trajectory and calculate the curvature by substituting these data into the curvature equation of equation (1).
[0043] Figure 6 is a diagram illustrating the direction of mouse movement. Feature calculation units 143-1 to 143-4 extract mouse features based on mouse movement data recorded in the mouse log (left-click data, right-click data, and the eight directions of mouse movement based on the coordinates of the cursor and the rotation amount of the up / down wheel (Figure 6)). The features are velocity, initial velocity, and acceleration for each direction. Additionally, the features include velocity, initial velocity, and acceleration for each distance the mouse moves.
[0044] The feature calculation unit 143-1 outputs the calculated features to the first estimation unit 144-1. The feature calculation unit 143-2 outputs the calculated features to the second estimation unit 144-2. The feature calculation unit 143-3 outputs the calculated features to the third estimation unit 144-3. The feature calculation unit 143-4 outputs the calculated features to the fourth estimation unit 144-4.
[0045] The first estimation unit 144-1, the second estimation unit 144-2, the third estimation unit 144-3, and the fourth estimation unit 144-4 authenticate whether the mouse operations based on the input mouse log are those of the first user, based on the input features.
[0046] The first estimation unit 144-1 has a first learning model 145-1. The first learning model 145-1 is a model that has learned features based on the mouse operation characteristics of a first user while running application A of AW.
[0047] The second estimation unit 144-2 has a second learning model 145-2. The second learning model 145-2 is a model that has learned features based on the mouse operation characteristics of a first user while running application B of AW.
[0048] The third estimation unit 144-3 has a third learning model 145-3. The third learning model 145-3 is a model that has learned features based on the mouse operation characteristics of a first user while running application C of AW.
[0049] The fourth estimation unit 144-4 has a fourth learning model 145-4. The fourth learning model 145-4 is a model that has learned features based on the mouse operation characteristics of a first user while running application D of AW.
[0050] The first learning model 145-1, the second learning model 145-2, the third learning model 145-3, and the fourth learning model 145-4, when given feature input, output, for example, a numerical value for anomaly or a score that represents the individual's resemblance. For example, the anomaly score is the degree to which the characteristics of the input mouse information deviate from the mouse characteristics of the first user, and a higher value indicates a greater deviation from the mouse characteristics of the first user.
[0051] The detection unit 146 detects, based on the estimation results from the estimation unit 142, whether the mouse operation based on the input mouse log is a mouse operation by the first user. If the detection unit 146 detects that the mouse operation is by the first user, it allows this user to operate the terminal device. If the detection unit 146 does not detect that the mouse operation is by the first user, it outputs an alert.
[0052] The detection unit 146 detects, for example, that the mouse operation is performed by the first user if the numerical value of the abnormality level output by the estimation unit 142 is less than a predetermined threshold, and permits the user to operate the terminal device. The detection unit 146 detects that the mouse operation is not performed by the user if the numerical value of the abnormality level output by the estimation unit 142 is equal to or greater than a predetermined threshold, and outputs an alert. The predetermined threshold may be set for each application, for example, or it may be set according to the statistical value of the output value of the learning model corresponding to each application. The predetermined threshold may be set in advance for each application, or it may be changed based on the authentication processing result by the authentication device 10.
[0053] [Learning Process] The learning process according to the embodiment will now be described. Figure 7 is a flowchart showing the processing steps of the learning process according to the embodiment. The first learning model 145-1 to the fourth learning model 145-4 learn feature quantities based on the mouse operation characteristics of the first user by executing each processing step shown in Figure 7 for each application.
[0054] The authentication device 10 accepts AW information and mouse operation information as input (steps S11, S13). Furthermore, the accuracy of mouse authentication is improved by assigning the first user the same authentication task, rather than a general-purpose task. The AW logger 11 generates an AW log based on the input AW information (step S12). The mouse logger 12 generates a mouse log based on the input mouse operation information (step S14).
[0055] The identification unit 141 performs identification processing based on the AW log to identify the type of AW application corresponding to the input mouse log (step S15).
[0056] Next, the estimation unit 142 uses a learning model corresponding to the application identified in step S15 to perform an estimation process to estimate whether the mouse operation based on the input mouse log is a mouse operation performed by the first user (step S16). The authentication device 10 extracts features using a feature extraction unit corresponding to the application identified in step S15, and inputs the extracted features into the learning model corresponding to the application identified in step S15 to obtain, for example, anomalies.
[0057] The learning unit 15 updates the parameters of the learning model that underwent estimation processing based on the estimation results estimated in step S16 (step S17). The learning unit 15 determines whether or not the learning termination conditions are met (step S18). The termination conditions include, for example, that the parameter updates for each learning model have reached a predetermined number of times, or that the amount of parameter updates has fallen below a predetermined threshold.
[0058] If the termination condition is not met (step S18: No), the authentication device 10 returns to steps S11 and S13 to continue the learning process and repeats each learning step. If the termination condition is met (step S18: Yes), the authentication device 10 terminates the learning process.
[0059] [Authentication process] Next, the authentication process according to the embodiment will be described. Figure 8 is a flowchart showing the processing procedure of the authentication process according to the embodiment.
[0060] The authentication device 10 accepts AW information and mouse operation information as input (steps S21, S23). The AW logger 11 generates an AW log based on the input AW information (step S22). The mouse logger 12 generates a mouse log based on the input mouse operation information (step S24).
[0061] The identification unit 141 performs an identification process based on the AW log to identify the type of AW application that corresponds to the input mouse log (step S25).
[0062] Next, the estimation unit 142 performs an estimation process to determine whether the mouse operations based on the input mouse log were performed by the first user (step S26). In step S27, the estimation unit 142 extracts features in the feature extraction unit corresponding to the application identified in step S26. The estimation unit 142 inputs the extracted features into the learning model corresponding to the application identified in step S26 to obtain an estimation result (e.g., anomaly score).
[0063] Based on the estimation result in step S27, the detection unit 146 detects whether the mouse operation based on the input log was performed by the first user (step S27).
[0064] If the detection unit 146 detects that the mouse operation is performed by the first user (step S27: Yes), it allows the user to operate the terminal device. If the detection unit 146 detects that the mouse operation is not performed by the first user (step S27: No), it outputs an alert (step S28).
[0065] [Effects of the embodiment] Figure 9 illustrates conventional mouse authentication. Conventional mouse authentication used a general-purpose learning model to perform authentication from input mouse logs, but it had the problem of unstable authentication accuracy.
[0066] Figures 10 and 11 illustrate a mouse authentication method according to an embodiment. The authentication device 10 acquires the mouse log of the first user along with the AW log and constructs a learning model for each application information during mouse operation. Specifically, the authentication device 10 trains each learning model with features based on the mouse operation characteristics of the first user while the corresponding application is running. Therefore, in this embodiment, the estimation accuracy of the learning model can be improved by enabling the learning of features for each application.
[0067] Then, during authentication, the authentication device 10 acquires AW information along with mouse operation information and identifies that the application being operated is "Word". Subsequently, the authentication device 10 inputs the mouse log features (arrow Y11 in Figure 10) into a learning model corresponding to the identified application type (for example, the first learning model 145-1) and estimates the degree of abnormality.
[0068] Then, if the numerical value of abnormality "0.2" (arrow Y12 in Figure 11) output by the first learning model 145-1 is less than a predetermined threshold, the authentication device 10 detects that the mouse operation is performed by the first user and permits the user to operate the terminal device.
[0069] In this way, the authentication device 10 acquires the AW log along with the mouse log of the first user and identifies the type of application corresponding to the mouse log. Then, the authentication device 10 performs authentication based on mouse operations using a learning model corresponding to the identified type of application, enabling highly accurate estimation and thus improving the accuracy of authentication based on mouse operations.
[0070] [Differentiation] Next, a modified example of the embodiment will be described. Figure 12 is a diagram showing an example of an authentication device configuration according to a modified example of the embodiment.
[0071] As shown in Figure 12, the authentication device 210 is connected to the first input unit 20, the second input unit 30, and the third input unit 40.
[0072] The third input unit 40 is, for example, the keyboard of a terminal device. The third input unit 40 inputs keyboard operation information from the operator of the terminal device to the authentication device 10.
[0073] Operation information input from the third input unit 40 is input to the control unit 13 via a keyboard driver (not shown). Figure 13 shows an image of key input. When a user presses a key on the keyboard, a scan code based on the physical placement on the keyboard is passed to the Keyboard device driver. The Keyboard device driver creates a message along with a virtual key code mapped by the driver and various information about the typing (for example, how many times the same key has been pressed consecutively). The active application receives this message and the key is entered. The keyboard driver monitors this message and outputs information about the key operation.
[0074] The authentication device 210 has a control unit 213 that includes an authentication unit 214 and a learning unit 215, instead of the control unit 13 shown in Figure 2.
[0075] The authentication unit 214 uses the fifth learning model 2149-1 (keyboard learning model) (described later) to estimate whether the UI operation based on the input UI operation information is a UI operation performed by the first user, based on the keyboard operation information input along with the mouse operation information as UI operation information.
[0076] In addition to the functions of the learning unit 15, the learning unit 215 constructs a fifth learning model 2149-1), which is a machine learning model, and trains this learning model to learn features based on the keyboard operations of the first user. The learning unit 215 uses the keyboard operation information performed by the first user to train the fifth learning model 2149-1 to learn features based on the keyboard operation characteristics of the first user.
[0077] [Authentication Department] Figure 14 shows the configuration of the authentication unit 214 shown in Figure 12. As shown in Figure 14, the authentication unit 214 has an additional estimation unit 2147 (second estimation unit) compared to the authentication unit 14 shown in Figure 3. The authentication unit 214 has a detection unit 2146 instead of the detection unit 146 shown in Figure 3.
[0078] The estimation unit 2147 uses a fifth learning model 2149-1 to estimate whether the keyboard operation based on the input keyboard operation information is a keyboard operation performed by a first user, based on the keyboard operation information input along with the mouse operation information as UI operation information. The estimation unit 2147 includes a feature calculation unit 2148 and a fifth estimation unit 2149.
[0079] The feature calculation unit 2148 calculates features based on the input keyboard operation information. The feature calculation unit 2148 acquires a series of keystroke information indicating the typing to be detected as keyboard operation information, and vectorizes the acquired series of keystroke information. The keystroke information is information that indicates, for example, the key pressed during typing, the time the key was pressed (Down), and the time the key was released (Up) (see Figure 15).
[0080] The fifth estimation unit 2149 has a fifth learning model 2149-1. The fifth learning model 2149-1 is a model that has learned features based on the typing characteristics of the first user. When features are input from the feature calculation unit 2148, the fifth learning model 2149-1 outputs, for example, a numerical value for abnormality or a score that represents the likelihood of it being the user.
[0081] The detection unit 2146 detects whether the UI operation based on the input UI operation information is performed by a first user, based on the estimation results from the estimation unit 142 and the estimation results from the estimation unit 2147. If the detection unit 2146 detects that the UI operation is performed by a first user, it permits this user to operate the terminal device. If the UI operation is not performed by a first user, the detection unit 2146 outputs an alert.
[0082] The detection unit 2146 performs a calculation by substituting, for example, the numerical value of the abnormality output by the estimation unit 142 and the abnormality output by the estimation unit 2147 into a predetermined calculation formula. If the calculated value is less than a predetermined threshold, the detection unit 2146 detects that it is a UI operation by the first user and permits the user to operate the terminal device. The calculation formula is an addition or multiplication formula of the numerical value of the abnormality output by the estimation unit 142 and the abnormality output by the estimation unit 2147, and may also be a formula in which the output values of the estimation unit 142 and / or the output values of the estimation unit 2147 are weighted.
[0083] [Learning Process] The learning unit 215 learns a characteristic model of the first user's typing through unsupervised learning using the first user's free typing. For example, the learning unit 215 learns a machine learning model to minimize the anomaly score for the keystroke information of the first user's free typing through unsupervised machine learning on the keystroke information of the first user's free typing.
[0084] Figure 16 is a flowchart showing the processing procedure of the learning process according to a modified embodiment. Figure 16 describes the learning of the fifth learning model 2149-1. The first learning models 145-1 to the fourth learning models 134-4 learn feature quantities based on the mouse operation characteristics of the first user by executing each processing procedure shown in Figure 7 for each application.
[0085] The authentication device 210 obtains a series of keystroke information (keyboard operation information) indicating the content of the typing from the free typing of the first user (step S211).
[0086] The feature calculation unit 2148 vectorizes the series of keystroke information obtained in step S211 (step S212). For example, as shown in Figure 15, the feature calculation unit 2148 creates a sequence of numbers that includes the identification information of key1 and key2 from the Down and Up states of two adjacent keys (key1 and key2) in the series of keystroke information, as well as the duration for which key1 and key2 were pressed (Hold1 and Hold2), DD (Down to Down), UD (Up to Down), etc. as elements. The feature calculation unit 2148 performs the above process for N keys to create an N × (information amount) vector.
[0087] The learning unit 215 learns a model of the first user's typing characteristics by performing unsupervised machine learning on a series of keystroke information vectorized in step S212 (step S213). The learning unit 215 uses multiple N×(information amount) vectors created by the feature calculation unit 2148 as training data to perform machine learning on the model of the first user's typing characteristics.
[0088] This machine learning method performs anomaly detection based on keystrokes by learning to minimize the degree of anomaly relative to the user's own keystrokes. This machine learning method is classified as unsupervised learning because, for example, it only uses the typing data of the first user. The learning unit 215 may also perform supervised learning in the machine learning of the second user's typing characteristic model described above.
[0089] The learning unit 215 repeatedly learns the typing characteristic model described above, and for example, when the degree of abnormality stops changing, it terminates the learning process and sets the typing characteristic model at that time as the fifth learning model 2149-1 corresponding to the first user.
[0090] Furthermore, due to the above learning process, the anomaly score output by the typing characteristics model is minimized for the user's own keystrokes, but not for those of others. Therefore, when someone else's keystrokes are input into the trained typing characteristics model, a higher anomaly score is output compared to when the user's own keystrokes are input.
[0091] [Authentication process] Next, the authentication process relating to a modified embodiment will be described. Figure 17 is a flowchart showing the processing procedure for the authentication process relating to a modified embodiment.
[0092] Steps S221 to S225 in Figure 17 are the same processes as steps S21 to S25 in Figure 8. The first estimation process (step S226) in Figure 17 is the same process as step S26 in Figure 8.
[0093] In the authentication device 210, the estimation unit 2147 receives input of keyboard operation information (step S227) and performs a second estimation process (step S228).
[0094] In step S228, the estimation unit 2147 extracts features (for example, a vector of N × (information amount)) from the keyboard operation information. The estimation unit 2147 inputs the extracted features into the fifth learning model and obtains the estimation result (for example, the anomaly score).
[0095] Based on the estimation results in step S226 and step S228, the detection unit 2146 detects whether the input UI operation is a UI operation performed by the first user (step S229).
[0096] If the detection unit 2146 detects that the UI operation is performed by the first user (step S229: Yes), it allows the user to operate the terminal device. If the detection unit 146 detects that the UI operation is not performed by the first user (step S229: No), it outputs an alert (step S230).
[0097] In this way, the authentication device 210 uses keyboard operation information in addition to AW information and mouse operation information to estimate whether the UI operation is being performed by the first user, thereby further improving the accuracy of authentication.
[0098] [System configuration of the embodiment] Each component of the authentication device 10 is a functional concept and does not necessarily need to be physically configured as shown in the diagram. In other words, the specific forms of distribution and integration of the functions of the authentication device 10 are not limited to those shown in the diagram, and all or part of it can be configured by functionally or physically distributing or integrating it in any unit according to various loads and usage conditions.
[0099] Furthermore, each process performed in the authentication device 10 may be implemented, in whole or in part, by a CPU, a GPU (Graphics Processing Unit), and a program that is analyzed and executed by the CPU and GPU. Alternatively, each process performed in the authentication device 10 may be implemented as hardware using wired logic.
[0100] Furthermore, among the processes described in the embodiments, all or part of the processes described as being performed automatically can be performed manually. Alternatively, all or part of the processes described as being performed manually can be performed automatically by known methods. In addition, the processing procedures, control procedures, specific names, and information including various data and parameters described above and illustrated may be changed as appropriate unless otherwise specified.
[0101] [program] Figure 18 shows an example of a computer in which an authentication device 10 is realized when a program is executed. The computer 1000 has, for example, memory 1010 and a CPU 1020. The computer 1000 also has a hard disk drive interface 1030, a disk drive interface 1040, a serial port interface 1050, a video adapter 1060, and a network interface 1070. These components are connected by a bus 1080.
[0102] Memory 1010 includes ROM 1011 and RAM 1012. ROM 1011 stores, for example, a boot program such as the BIOS (Basic Input Output System). The hard disk drive interface 1030 is connected to the hard disk drive 1090. The disk drive interface 1040 is connected to the disk drive 1100. For example, a removable storage medium such as a magnetic disk or optical disk is inserted into the disk drive 1100. The serial port interface 1050 is connected to, for example, the mouse 1110 and the keyboard 1120. The video adapter 1060 is connected to, for example, the display 1130.
[0103] The hard disk drive 1090 stores, for example, an OS (Operating System) 1091, an application program 1092, a program module 1093, and program data 1094. That is, the program that defines each process of the authentication device 10 is implemented as a program module 1093 in which code executable by the computer 1000 is written. The program module 1093 is stored, for example, in the hard disk drive 1090. For example, a program module 1093 for performing the same processes as the functional configuration of the authentication device 10 is stored in the hard disk drive 1090. Note that the hard disk drive 1090 may be replaced by an SSD (Solid State Drive).
[0104] Furthermore, the configuration data used in the processing of the above-described embodiment is stored as program data 1094 in, for example, memory 1010 or hard disk drive 1090. The CPU 1020 then reads the program module 1093 and program data 1094 stored in memory 1010 or hard disk drive 1090 into RAM 1012 as needed and executes them.
[0105] Furthermore, the program module 1093 and program data 1094 are not limited to being stored in the hard disk drive 1090; for example, they may be stored in a removable storage medium and read by the CPU 1020 via a disk drive 1100 or the like. Alternatively, the program module 1093 and program data 1094 may be stored in another computer connected via a network (LAN (Local Area Network), WAN (Wide Area Network), etc.). The program module 1093 and program data 1094 may then be read by the CPU 1020 from the other computer via a network interface 1070.
[0106] The functions of the elements disclosed herein may be implemented using circuits or processing circuitry that include general-purpose processors, special-purpose processors, integrated circuits, ASICs (Application Specific Integrated Circuits), FPGAs (Field Programmable Gate Arrays), conventional circuits, and / or combinations thereof that are programmed using one or more programs stored in one or more memories, or otherwise configured to perform the disclosed functions. A processor is considered processing circuitry or circuitry because it includes transistors and other circuits. A processor may be a programmed processor that executes programs stored in memory. In this disclosure, a circuit, unit, or means is hardware that performs the enumerated functions, or hardware programmed to perform the enumerated functions. Hardware may be any hardware disclosed herein that is programmed or configured to perform the enumerated functions.
[0107] There is memory for storing a computer program that includes computer instructions. These computer instructions provide logic and routines that enable hardware (e.g., processing circuitry or circuitry) to perform the methods disclosed herein. This computer program can be implemented in commonly known forms, such as computer-readable storage media, computer program products, memory devices, recording media such as CD-ROMs and DVDs, and / or memory of FPGAs and ASICs.
[0108] Although embodiments applying the invention made by the present inventors have been described above, the present invention is not limited by the descriptions and drawings that constitute part of the disclosure of the present invention in these embodiments. That is, all other embodiments, examples, and operational techniques made by those skilled in the art based on these embodiments are included in the scope of the present invention. [Explanation of symbols]
[0109] 10,210 Authentication devices 11 AW Logger 12 Mouse Loggers 13,213 Control Unit 14,214 Authentication Department 15,215,2135 Learning Department 20 First input section 30 Second input section 40 Third input section 141 Identification Unit 142,2147 Estimation part 143-1, 143-2, 143-3, 143-4, 2148 Feature calculation unit 144-1 First Estimation Section 144-2 Second Estimation Section 144-3 Third Estimation Section 144-4 The fourth presumptive section 146,2146 detection unit 145-1 The First Learning Model 145-2 The Second Learning Model 145-3 The Third Learning Model 145-4 The Fourth Learning Model 2149 Fifth Estimation Section 2149-1 The fifth learning model
Claims
1. The learning unit constructs a machine learning model for each type of application during UI (User Interface) operation, and trains each learning model with features based on the UI operation characteristics of the first user during the execution of the corresponding application. An authentication unit identifies which application the UI operation based on the input UI operation information is for, and uses the learning model corresponding to the identified application to authenticate whether the UI operation based on the input UI operation information is a UI operation by the first user. An authentication device characterized by having the following features.
2. The authentication device according to claim 1, characterized in that the learning unit causes the learning model corresponding to the application of the active window to learn feature quantities based on the mouse operation characteristics while the application of the active window is running, based on the mouse operation information operated by the first user and the log of the active window when the first user operates the mouse.
3. The authentication unit, Based on the log of the active window during mouse operation, which is input along with the UI operation information, an identification unit identifies the type of application of the active window corresponding to the input mouse operation information. A first estimation unit uses a learning model corresponding to the application identified by the identification unit to estimate whether the mouse operation based on the input mouse operation information is a mouse operation by the first user. A detection unit detects whether the mouse operation based on the input mouse operation information is a mouse operation performed by the first user, based on the estimation result by the first estimation unit. The authentication device according to claim 2, characterized by having the following features.
4. The learning unit uses the keyboard operation information of the first user to train the keyboard learning model to learn feature quantities based on the keyboard operation characteristics of the first user. The authentication unit, A second estimation unit uses the keyboard operation information, which is input along with the mouse operation information as UI operation information, to estimate whether the keyboard operation based on the input keyboard operation information is a keyboard operation performed by the first user, using the keyboard learning model. It further possesses, The authentication device according to claim 3, characterized in that the detection unit detects whether or not a UI operation based on input UI operation information is a UI operation performed by the first user, based on the estimation result from the first estimation unit and the estimation result from the second estimation unit.
