Automatic data labeling method and related computer system for automatic data labeling

Abstract

An automatic data labeling method includes inputting a time series dataset; performing a subject labeling procedure on each value of the time series dataset based on at least one behavioral characteristic of a specified user behavior; and determining an optimal labeling combination corresponding to each value of the time series dataset and a disorder rating value based on a number of scenarios.

Description

Technical Field

[0001] This invention relates to an automatic data labeling method and a computer system for automatic data labeling, and more particularly to an automatic data labeling method and a computer system for automatic data labeling with specifyable characteristics. Prior Technology

[0002] User behavior analysis aims to understand user (e.g., customer or system) behavior by analyzing data related to user behavior to understand user needs. Existing methods for analyzing user behavior patterns include collecting time-series data and extracting data related to user behavior. However, these processes often cannot determine the starting point of user behavior, the duration of user behavior, or the numerical range of data generated by user behavior.

[0003] To optimize the user experience of laptops, conventional technology typically involves manually monitoring the temperature of key components to collect data on the performance of these components during user use, such as power consumption and temperature changes of the central processing unit (CPU) and graphics processing unit (GPU).

[0004] However, the power consumption and temperature changes of the main components of a laptop when the user is using it or not cannot be directly distinguished. This requires lengthy observation periods of the laptop's heating duration (e.g., 70 minutes) and cooling duration (e.g., 60 minutes). Furthermore, when manually observing the temperature of the main components, the laptop's heating behavior must be continuously monitored until it reaches an equilibrium temperature, or its cooling behavior must be monitored until it reaches an equilibrium temperature before proceeding to the next scenario experiment. This results in inefficient user behavior analysis.

[0005] Therefore, there is a need to improve the existing technology. Summary of the Invention

[0006] Therefore, the present invention provides an automatic data labeling method and a computer system for automatically labeling data, so as to automatically label a large dataset according to specified characteristics.

[0007] This invention discloses an automatic data labeling method, comprising: inputting a time series dataset; performing a subject labeling procedure on each value of the time series dataset based on at least one behavioral characteristic of a specified user behavior; and determining an optimal labeling combination corresponding to each value of the time series dataset and a disorder rating value based on a number of scenarios.

[0008] This invention also discloses a computer system for automatically tagging data, comprising a processing device and a memory device coupled to the processing device for storing code to instruct the processing device to execute a process for checking automatically tagged data, wherein the process includes inputting a time series dataset; performing a subject tagging procedure on each value of the time series dataset based on at least one behavioral characteristic of a specified user behavior; and determining an optimal tagging combination corresponding to each value of the time series dataset and a disorder rating value based on a number of scenarios. Simple Explanation of the Diagram

[0009] Figure 1 is a schematic diagram of a computer system according to one embodiment of the present invention. Figure 2 is a schematic diagram of an automatic data labeling process according to one embodiment of the present invention. Figure 3 is a schematic diagram of a main period marking method for a time series data set according to an embodiment of the present invention. Figure 4 is a schematic diagram of the main period marking method of the time series data set according to an embodiment of the present invention. Figures 5, 6, 7, and 8 are schematic diagrams illustrating the main period marking of a time-series data set according to an embodiment of the present invention. Figure 9 is a schematic diagram of a method for adjusting the main period marker of a time series data set according to an embodiment of the present invention. Figure 10 is a schematic diagram of the main period marking after adjusting the time series data set according to an embodiment of the present invention. Figure 11 is a schematic diagram of a main period marking method for time series data sets according to an embodiment of the present invention. Figures 12, 13, and 14 are schematic diagrams illustrating the main period marking of a time-series dataset according to an embodiment of the present invention. Figure 15 is a schematic diagram of the dynamic tracking time-series data flow according to one embodiment of the present invention. Implementation

[0010] Please refer to Figure 1, which is a schematic diagram of a computer system 10 according to one embodiment of the present invention. The computer system 10 is used for automatically tagging data and includes a processing device 102 and a memory device 104. The memory device 104 is coupled to the processing device 102 and is used to store code to instruct the processing device 102 to execute an automatic tagging data inspection process 20.

[0011] Please refer to Figure 2, which is a schematic diagram of the automatic data labeling process 20 according to an embodiment of the present invention. In one embodiment, the automatic data labeling process 20 can be used to label a time-series dataset based on at least one behavioral characteristic of a specified user behavior, wherein the time-series dataset contains different values. The automatic data labeling process 20 includes the following steps:

[0012] Step 202: Begin;

[0013] Step 204: Input the time series data set;

[0014] Step 206: Based on at least one behavioral characteristic of the specified user behavior, perform a subject labeling procedure on each value in the time series dataset;

[0015] Step 208: Based on a scenario quantity, determine the optimal combination of labels and a disorder rating value corresponding to each value in the time series dataset;

[0016] Step 210: End.

[0017] Automatic data labeling process 20 can automatically label a time-series dataset based on given specified user behavior. In step 204, the computer system 10 of this embodiment can collect time-series datasets about computer components based on the time-series datasets provided by the user or under different experimental scenarios. In one embodiment, the computer system 10 of this embodiment automatically labels the heat dissipation time-series data of a notebook computer hardware during operation, wherein the notebook computer hardware can be the power consumption and temperature changes of a central processing unit (CPU) or a graphics processing unit (GPU), and the data in the time-series dataset is the wattage or temperature value of the CPU or GPU. In the following embodiments, the temperature change of the heat-generating components of a notebook computer is used as an example, but this is not a limitation; desktop computers or other computer hardware components can also be applied to this invention.

[0018] Step 206 further performs a subject labeling procedure on each value in the time series dataset based on at least one behavioral characteristic of the specified user behavior. That is, the computer system 10 of this embodiment can perform a subject labeling procedure on the time series dataset based on at least one behavioral characteristic of the specified user behavior. In step 208, based on the number of scenarios (the number of subject labeling procedures that match the behavioral characteristics of the specified user behavior), the optimal labeling combination and disorder assessment value corresponding to each value in the time series dataset are determined. In this way, the computer system 10 of this embodiment can analyze user behavior based on the optimal labeling combination and disorder assessment value of the time series dataset.

[0019] Observing the timing data of the heat dissipation of notebook computers usually requires collecting data characteristics over a period of time. Therefore, the computer system 10 of this embodiment can mark the timing data according to the specified user behavior, and then use it for user behavior analysis.

[0020] The time-series dataset of this invention may include the following features:

[0021] 1. The labeled time-series data is a range;

[0022] 2. The tagged time-series data set must at least contain characteristics of the specified user behavior of the time-series data;

[0023] 3. The time series data that are labeled have the same interval length.

[0024] Specifying user behavior is a way to describe user behavior, such as the characteristics of a laptop's heat dissipation test, and effectively label the data generated by user behavior (or laptop heat dissipation test).

[0025] Taking user behavior of a laptop as an example, when a user starts using the laptop, the laptop's graphics processor temperature tends to rise; conversely, when the user stops using the laptop, the laptop's graphics processor temperature tends to fall. Therefore, the computer system 10 of this embodiment uses the characteristic of the laptop's graphics processor temperature starting to rise or begin to fall as the designated characteristic of the tag data.

[0026] In another embodiment, the temperature of the graphics processor of the notebook computer can be used as a characteristic of specified user behavior when it starts to rise or fall to a stable temperature. Therefore, the computer system 10 of this embodiment can automatically tag relevant data for analyzing user behavior.

[0027] Assume a time series data set: ,in, It is the value of the nth time-series data at time t. The startup specified characteristic for user behavior is: ,in 1 indicates startup, 0 indicates startup not started. The representative observed 10 data points.

[0028] The complete specification feature for specifying user behavior is: ,in 1 indicates completion, 0 indicates incompleteness.

[0029] The disorder assessment for time-series datasets is as follows: ,in To specify the initial time series of user behavior, This is a time series of completed user behaviors, where a higher disorder rating indicates greater disorder.

[0030] When a time series data set is input, the computer system 10 of this embodiment of the invention performs subject marking on each value of the time series data according to the specified characteristics of the specified user behavior, namely the initiation specified characteristics and the completion specified characteristics. Each marked interval is a subject period (principal period) of the specified user behavior.

[0031] Since the timing data set consists of the temperature or wattage values ​​of the laptop's components, activating the specified characteristic means that the temperature (or wattage) of the laptop's heat-generating components starts to rise from an equilibrium temperature (or wattage), while completing the specified characteristic means that the temperature (or wattage) of the laptop's heat-generating components starts to fall from a high temperature (or wattage).

[0032] Therefore, the specified characteristic for activation is given by equation (1): …(1)

[0033] in ,and , This represents the range of temperature change.

[0034] Similarly, the method to complete the specified characteristic is Equation (2): …(2)

[0035] in ,and .

[0036] The computer system 10 of this embodiment of the invention is... and Then, all start times can be determined. and completion time The calculation method is as follows:

[0037] …(3)

[0038] …(4)

[0039] …(5)

[0040] It is worth noting that if a time series does not exist If , it means that the time series cannot be determined to have the behavior of the specified user.

[0041] In determining the number of scenarios that a time-series data set should include. Subsequently, the computer system 10 of this embodiment can mark the subject period of a specified user behavior for any time series dataset. Please refer to Figure 3, which is a schematic diagram of a subject period marking method 30 for a time series dataset according to an embodiment of this invention. The subject period marking method 30 includes the following steps:

[0042] Step 302: Begin;

[0043] Step 304: Find the candidate set ;

[0044] Step 306: Confirm the candidate set Is it empty? If yes, proceed to step 308; otherwise, proceed to step 310.

[0045] Step 308: Assume the optimal label combination is empty, the corresponding disorder rating value is empty, and the corresponding number of scenarios is empty;

[0046] Step 310: Find the optimal combination of markers ;

[0047] Step 312: Calculate the randomness score corresponding to the optimal label combination. ;

[0048] Step 314: Output the optimal combination of tags Corresponding disorder rating values and the number of corresponding situations ;

[0049] Step 316: Completed.

[0050] Main phase marking method 30 is used to determine the candidate set. (Step 304), and confirm whether there is a matching optimal combination of tags in the candidate set (Step 306), so as to be in the candidate set When not empty, determine the optimal combination of tags. (Step 310), and the corresponding disorder assessment value (Step 312).

[0051] It is worth noting the number of scenarios Usually with tag combinations The value increases and decreases, therefore, the computer system 10 of this embodiment can obtain the candidate set using a general binary search method combined with a local greedy search method. .

[0052] In addition, any The number of corresponding scenarios All are the same. Therefore, the number of scenarios It is a combination of tags A constant that changes, corresponding to a randomness assessment value. The system is based on all combinations of markers. What was obtained.

[0053] Taking the heating element of a laptop computer as an example, when the ratio of the time required to reach the equilibrium temperature in each situation to the time required to recover from the heating temperature to the equilibrium temperature approaches a certain value. In one embodiment, the disorder assessment numerical method is as follows: …(6)

[0054] in, , , …(7)

[0055] in, This represents the ratio of cooling to heating.

[0056] Please refer to Figure 4, which is a schematic diagram of a subject period marking method 40 for a timing dataset according to an embodiment of the present invention. The subject period marking method 40 includes the following steps:

[0057] Step 402: Input a time series data set ;

[0058] Step 404: For each time series data in the time series data set Output the corresponding optimal combination of tags. Corresponding disorder rating values and the number of corresponding situations ;

[0059] Step 406: Confirm whether the number of scenarios corresponding to each time series data is empty. If yes, proceed to step 414; otherwise, proceed to step 408.

[0060] Step 408: Determine the candidate set

[0061] Step 410: Determine the best combination of markers ;

[0062] Step 412: Combine with tags This corresponds to the start time sequence of user behavior. Time sequence of user behavior completion All time-series data are labeled;

[0063] Step 414: Complete.

[0064] In step 404 of the main period marking method 40, the time series data set input in step 402 is used... Each time-series data point determines the optimal combination of labels for the output. Corresponding disorder rating values and the number of corresponding situations In step 406, it is confirmed whether the number of scenarios is empty, and then in step 408, the candidate set is determined. and in step 410 determine the optimal combination of markers. This allows the computer system 10 to mark the timing data in step 412.

[0065] Figures 5, 6, 7, and 8 are schematic diagrams illustrating the marking of the main period of a time series dataset according to an embodiment of the present invention. The horizontal axis represents the number of units in the time series dataset, the vertical axis represents the numerical value (temperature) of the time series dataset, the dashed line represents the marked time series data, and the solid line represents the unmarked time series data.

[0066] Because the time-series data marked as the main period may have a delay, i.e., the marked start time or finish time is after the actual start time or finish time, the computer system 10 of this embodiment further adjusts the time-series data of the marked main period. Please refer to Figure 9, which is a schematic diagram of a main period marking adjustment method 90 of a time-series dataset according to an embodiment of this invention. The main period marking adjustment method 90 includes the following steps:

[0067] Step 902: Assumption and ;

[0068] Step 904: Calculation and ;

[0069] Step 906: For each subject period Calculate the main period of correction ,in (If it does not exist, then let) ) (If it does not exist, then let) )

[0070] Step 908: During each amendment period Re-label the time series data;

[0071] Step 910: Complete.

[0072] The main period marker adjustment method 90 is in and Under the assumption (step 902), calculate in step 904. and During step 906, the main body is modified. Then, the time series data is re-labeled (step 908).

[0073] Please refer to Figure 10, which is a schematic diagram of the main period marking after adjusting the time series data set according to an embodiment of the present invention. The horizontal axis represents the number of units in the time series data set, the vertical axis represents the numerical value (temperature) of the time series data set, the dashed line represents the marked time series data, and the solid line represents the unmarked time series data.

[0074] Compared to Figure 5, the start and end times of the time series data set marked in Figure 10 are closer to the actual start or end times.

[0075] The computer system 10 of this embodiment can further integrate the subject period marking method 40 and the subject period marking adjustment method 90 into the time series dataset, so that the time series dataset can still maintain the marking characteristics. Please refer to Figure 11, which is a schematic diagram of a subject period marking method 1100 for a time series dataset according to an embodiment of this invention. The subject period marking method 1100 includes the following steps:

[0076] Step 1102: Find the optimal combination of labels ;

[0077] Step 1104: Output Corresponding time-series data and the start time series of user behavior and the time sequence of user behavior completion ;

[0078] Step 1106: Label the time series data from Step 1104 with start time series and finish time series;

[0079] Step 1108: Adjust the marked time series data from step 1106 to obtain the corrected starting time series. and the corrected completion time series ;

[0080] Step 1110: Using the adjusted time series from step 1108, label each time series data point;

[0081] Step 1112: Completed.

[0082] The main phase marking method 1100 can then use the optimal marking combination from step 1102. Then, based on the time series data from step 1104 and the starting time sequence of user behavior... and the time sequence of user behavior completion The starting time series and the ending time series of the time series data are marked accordingly (step 1106) and the time series data is adjusted (step 1108), and then the adjusted marking procedure is completed in step 1110.

[0083] Please refer to Figures 12, 13, and 14. Figures 12, 13, and 14 are schematic diagrams of marking the main period of a time series data set according to an embodiment of the present invention. The horizontal axis represents the number of units in the time series data set, the vertical axis represents the numerical value (temperature) of the time series data set, the dashed line represents the marked time series data, and the solid line represents the unmarked time series data.

[0084] In other embodiments, the start and end times of the time series data sets marked in Figures 12, 13, and 14 are closer to the actual start or end times.

[0085] In addition, to achieve fully automated usage scenarios, the computer system 10 of this embodiment may further include a function for dynamically tracking timing data, so as to dynamically issue a start signal or a completion signal after the timing data set is generated to mark the timing data. Please refer to Figure 15, which is a schematic diagram of a dynamic timing data tracking process 1500 according to one embodiment of this invention. The dynamic timing data tracking process 1500 includes the following steps:

[0086] Step 1502: Determine the parameter P for the time series data set to be notified;

[0087] Step 1504: Specify the timing data for detection;

[0088] Step 1506: Decision ;

[0089] Step 1508: Calculation ;

[0090] Step 1510: Confirm If yes, proceed to step 1512; otherwise, proceed to step 1508.

[0091] Step 1512: Send a start signal;

[0092] Step 1514: Calculation ;

[0093] Step 1516: Confirm If yes, proceed to step 1518; otherwise, proceed to step 1514.

[0094] Step 1518: Send a shutdown signal.

[0095] The dynamic tracking timing data flow 1500 can confirm whether to issue a start signal after confirming that the number of consecutive start signals is greater than parameter P (steps 1508, 1510, 1512), and confirm whether to issue a stop signal after confirming that the number of consecutive completion signals is greater than parameter P (steps 1514, 1516, 1518).

[0096] Taking Figures 10, 12, 13, and 14 as examples, after the number of units in the time series dataset reaches 10,000, there is no start signal (the time series dataset is not marked with a start time), and the time series data is also in an idle state. Therefore, the computer system 10 of this embodiment of the invention can automatically end the above-mentioned automatic marking process when the number of experimental scenarios to be tested or the latest start time interval is known.

[0097] It is worth noting that the larger the parameter P regarding the time series dataset, the later the notification time will be. Furthermore, the combination of time series dataset and label... This can be obtained through the above-described marking method. Therefore, the dynamic tracking time-series data flow 1500 of this embodiment of the invention can obtain data from a given time-series data set and parameters. Under these conditions, time-series data is dynamically tracked and notifications are automatically issued to achieve a fully automated effect.

[0098] The above embodiments describe the concept of the present invention. Those skilled in the art can make appropriate modifications accordingly and are not limited thereto. They are not limited to the above examples, but can be adjusted according to the needs of users or data users, and all of them fall within the scope of the present invention.

[0099] In summary, the embodiments of the present invention provide an automatic data labeling method and a computer system for automatically labeling data, thereby efficiently and automatically labeling a large amount of time-series data to achieve the purpose of analyzing user behavior and thus producing products that better suit users. The above description is only a preferred embodiment of the present invention. All equivalent changes and modifications made in accordance with the claims of the present invention shall be covered by the present invention.

[0100] 10: Computer System 102: Processing device 104: Memory Device 20: Automatic Data Tagging Process 202,204,206,208,210: Steps 30: Main Period Marking Method 302, 304, 306, 308, 310, 312, 314, 316: Steps 40: Main Period Marking Method 402, 404, 406, 408, 410, 412, 414: Steps 90: Main Period Marking Adjustment Method 902, 904, 906, 908, 910: Steps 1100: Main Period Marking Method 1102, 1104, 1106, 1108, 1110, 1112: Steps 1500: Dynamic Tracking of Time Series Data Flow 1502, 1504, 1506, 1508, 1510, 1512, 1514, 1516, 1518: Steps

Claims

1. An automatic data labeling method for a computer system, comprising: inputting a time-series dataset; performing a subject labeling procedure on each value of the time-series dataset based on at least one behavioral characteristic of a specified user behavior; and determining an optimal labeling combination and a randomness assessment value corresponding to each value of the time-series dataset based on a number of scenarios; calculating a start count or a completion count of the time-series dataset and determining a shutdown signal; wherein each subject labeling interval of the subject labeling procedure conforms to at least one behavioral characteristic of the specified user behavior; wherein the number of scenarios is obtained based on a binary search method and a greedy search method; and wherein each subject labeling interval of the subject labeling procedure has the same duration.

2. The automatic data labeling method as described in claim 1, wherein the main labeling procedure labels each value of the time series dataset according to an initiation specified characteristic and a completion specified characteristic, and determines a candidate set according to the number of scenarios for the time series dataset; wherein the disorder rating value is determined according to the optimal labeling combination.

3. The automatic data labeling method as described in claim 1, further comprising: modifying each subject labeling interval of the subject labeling procedure to relabel the time series data set; and labeling the time series data set according to the optimal labeling combination and the modified subject labeling interval.

4. The automatic data labeling method as described in claim 1, further comprising: setting a dynamic tracking parameter, and determining to stop automatically labeling the time series data set based on the dynamic tracking parameter, a number of starts and a number of completions of the time series data set.

5. A computer system for automatically tagging data, comprising: a processing unit; and a memory unit coupled to the processing unit for storing code to instruct the processing unit to perform a process of checking automatically tagged data, wherein, The process includes: inputting a time-series dataset; performing a subject labeling procedure on each value in the time-series dataset based on at least one behavioral characteristic of a specified user behavior; determining an optimal labeling combination and a randomness assessment value corresponding to each value in the time-series dataset based on a number of scenarios; calculating a start count or a completion count in the time-series dataset to determine a shutdown signal; wherein each subject labeling interval of the subject labeling procedure conforms to at least one behavioral characteristic of the specified user behavior; wherein the number of scenarios is obtained using a binary search method and a greedy search method; wherein each subject labeling interval of the subject labeling procedure has the same duration.

6. The computer system for automatically labeling data as described in claim 5, wherein the main labeling program labels each value of the time series dataset according to a start-up specified characteristic and a completion specified characteristic, and determines a candidate set according to the number of scenarios for the time series dataset; wherein the disorder rating value is determined according to the optimal labeling combination.

7. The computer system for automatically labeling data as described in claim 5, wherein the automatic data labeling process further includes: modifying each subject labeling interval of the subject labeling procedure to relabel the time series data set; and labeling the time series data set according to the optimal labeling combination and the modified subject labeling interval.

8. The computer system for automatically labeling data as described in claim 5, wherein the automatic labeling process further includes: setting a dynamic tracking parameter, and determining to stop automatically labeling the time series data set based on the dynamic tracking parameter, a number of starts and a number of completions of the time series data set.

Images