Computing device capable of providing artificial intelligence-based recommendation function for recommending next action command by analyzing action command execution history of user, and operating method thereof

Abstract

The present invention provides a computing device capable of providing an artificial intelligence-based recommendation function for recommending a next action command by analyzing an action command execution history of a user, and an operating method thereof, and thus can support that the appropriateness of which action command to be executed next can be more easily determined when the user executes a series of action commands in the computing device.

Description

A computing device and its operating method capable of providing an artificial intelligence-based recommendation function that recommends the next action command by analyzing the user's action command execution history.

[0001] The present invention relates to a computing device and an operating method thereof that can provide an artificial intelligence-based recommendation function that recommends the next action command by analyzing the user's action command execution history.

[0002] Recently, with the development of artificial intelligence technology, artificial intelligence technology is being utilized in various fields.

[0003] For example, in the marketing field, there are technologies that use artificial intelligence to analyze customers' psychology and provide them with customized advertising information, or technologies that use artificial intelligence to predict climate change.

[0004] In this regard, as the use of computing devices such as desktop computers, laptops, and tablet computers has increased in recent years, the introduction of artificial intelligence technology to enhance user convenience in the use of these computing devices may be considered.

[0005] Specifically, a technology may be introduced that analyzes the history of which action commands a user has authorized on a computing device and recommends the next action command for the user. For example, if a user executes an action command to run a document creation program on a computing device and then executes an action command instructing the document creation program to run a browser to search for a document, a technology may be introduced that recommends to the user that the next action command be an action command instructing the user to search for a recently created document, since the user may want to search for a recently created document.

[0006] If such technology is introduced, it will be possible to improve the convenience of users of computing devices by reducing the time they spend thinking about which action command to execute next.

[0007] Prior art documents that can be referenced in relation to the present invention include "Ashish Vaswani et al., "Attention Is All You Need", arXiv:1706.03762, June 2017."

[0008] The present invention proposes a computing device and an operating method thereof that can provide an artificial intelligence-based recommendation function that recommends the next action command by analyzing the user's action command execution history, thereby enabling the user to more easily determine which action command is appropriate to execute next when the user executes a series of action commands on the computing device.

[0009] According to one embodiment of the present invention, a computing device capable of providing an artificial intelligence-based recommendation function that analyzes a user's action command execution history and recommends a next action command, comprises: a command database in which a plurality of preset action commands, each of the plurality of action commands meaning a command for calling an action designated as executable by the computing device, is stored; a download unit that downloads data on a command recommendation model from a preset model management system server when an activation command for an action command recommendation function is input from a user of the computing device, the command recommendation model being an artificial intelligence model that has been machine-learned in advance to select a next action command corresponding to a command set input from among the plurality of action commands when a command set consisting of k (k is a natural number greater than or equal to 2) action commands sequentially executed on the computing device is input; a judgment event generation unit that, when the data for the command recommendation model is downloaded, generates a judgment event for determining whether to execute the action command recommendation function whenever the cumulative number of action commands executed by the user reaches k in the computing device; and at a certain point in time, the judgment event generation unit that generates a judgment event for determining whether to execute the action command recommendation function. Through the event generating unit, when a judgment event occurs, within a preset waiting time from the time when the judgment event occurs, it is checked whether an additional action command by the user is executed on the computing device, and if it is confirmed that it is not executed, the generation of the judgment event through the judgment event generating unit is stopped, and then an execution event generating unit for executing an action command recommendation function is generated, through the execution event generating unit, when an execution event occurs, k first action commands that have been cumulatively executed on the computing device from the time when the previous judgment event occurred to the time when the judgment event occurred are checked, and the first action commands areThe method comprises: arranging the first command set according to the execution order, and then processing the first command set as an input to the command recommendation model so that one of the plurality of action commands is selected through the command recommendation model; and then, when one of the action commands is selected through the command recommendation model, a designation unit for designating the selected action command as a recommended action command; and a guidance unit for generating and displaying on a screen a guidance message guiding that the recommended action command is recommended as the next action command corresponding to the first action commands, and then processing so that the generation of a judgment event through the judgment event generation unit is resumed.

[0010] In addition, according to an embodiment of the present invention, a method for operating a computing device capable of providing an artificial intelligence-based recommendation function that recommends a next action command by analyzing a user's action command execution history comprises the steps of: maintaining a command database in which a plurality of preset action commands, each of the plurality of action commands meaning a command for calling an action designated as executable on the computing device, is stored; when an activation command for an action command recommendation function is received from a user of the computing device, downloading data on a command recommendation model from a preset model management system server, the command recommendation model being an artificial intelligence model that has been pre-machine-learned to select a next action command corresponding to a command set received as an input from among the plurality of action commands when a command set composed of k (k is a natural number greater than or equal to 2) action commands sequentially executed on the computing device is received as an input; when the data for the command recommendation model is downloaded, generating a determination event for determining whether to execute the action command recommendation function on the computing device whenever the cumulative number of action commands executed by the user reaches k; at a certain point in time, Through the step of generating the above judgment event, when the judgment event occurs, within a preset waiting time from the time when the judgment event occurs, a step of checking whether an additional action command by the user is executed on the computing device, and if it is confirmed that it is not executed, a step of processing to stop the generation of the judgment event, and then generating an execution event for executing an action command recommendation function, through the step of generating the execution event, when the execution event occurs, k first action commands that have been cumulatively executed on the computing device from the time when the previous judgment event occurred to the time when the judgment event occurred are checked, and the first action commands areThe method comprises the steps of: arranging the first command set according to the execution order, and then processing the first command set as an input to the command recommendation model so that one of the plurality of action commands is selected through the command recommendation model; and then, when one of the action commands is selected through the command recommendation model, designating the selected action command as a recommended action command; and generating and displaying on a screen a guidance message guiding that the recommended action command is recommended as the next action command corresponding to the first action commands, and then processing the occurrence of a judgment event that was interrupted so that it is resumed.

[0011] The present invention proposes a computing device and an operating method thereof that can provide an artificial intelligence-based recommendation function that recommends the next action command by analyzing the user's action command execution history, thereby enabling the user to more easily determine which action command is appropriate to execute next when the user executes a series of action commands on the computing device.

[0012] FIG. 1 is a diagram illustrating the structure of a computing device according to one embodiment of the present invention.

[0013] FIG. 2 is a flowchart illustrating an operation method of a computing device according to an embodiment of the present invention.

[0014] Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings. This description is not intended to limit the present invention to specific embodiments, but should be understood to include all modifications, equivalents, and substitutes included in the spirit and technical scope of the present invention. In describing each drawing, similar reference numerals are used to indicate similar components, and unless otherwise defined, all terms used in this specification, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the art to which the present invention belongs.

[0015] In this document, when a part is said to "include" a certain component, this does not mean that other components are excluded, but rather that other components may be included, unless otherwise specifically stated. In addition, in various embodiments of the present invention, each component, functional block, or means may be composed of one or more subcomponents, and the electrical, electronic, and mechanical functions performed by each component may be implemented by various known elements or mechanical elements, such as electronic circuits, integrated circuits, and ASICs (Application Specific Integrated Circuits), and each may be implemented separately, or two or more may be integrated into one.

[0016] Meanwhile, the blocks in the attached block diagram or the steps in the flowchart may be interpreted as computer program instructions that are loaded into the processor or memory of a data processing device such as a general-purpose computer, special-purpose computer, portable laptop computer, or network computer to perform specified functions. Since these computer program instructions can be stored in a memory equipped in the computer device or a computer-readable memory, the functions described in the blocks in the block diagram or the steps in the flowchart may be produced as a product that includes a command means for performing the same. In addition, each block or each step may represent a module, segment, or part of code that includes one or more executable instructions for performing a specific logical function(s). It should also be noted that in some alternative embodiments, the functions mentioned in the blocks or steps may be performed out of the specified order. For example, two blocks or steps depicted in succession may be performed substantially simultaneously or in reverse order, and in some cases, some blocks or steps may be performed with some blocks or steps omitted.

[0017] FIG. 1 is a diagram illustrating the structure of a computing device according to one embodiment of the present invention.

[0018] A computing device (110) according to the present invention is a device capable of providing an artificial intelligence-based recommendation function that recommends the next action command by analyzing the user's action command execution history, and includes a command database (111), a download unit (112), a judgment event generation unit (113), an execution event generation unit (114), a designation unit (115), and a guidance unit (116).

[0019] The computing device (110) of the present invention refers to a device having a predetermined microprocessor-based computing capability, such as a desktop computer, laptop, tablet computer, etc.

[0020] A plurality of preset action commands are stored in the command database (111).

[0021] Here, each of the plurality of action commands refers to a command for calling an action designated as executable on a computing device (110), and specifically, refers to an action command instructing to run a document creation program, an action command instructing to run a browser, an action command instructing to copy a file, an action command instructing to open a folder, and the like.

[0022] At this time, according to one embodiment of the present invention, the command database (111) may additionally store a preset embedding vector corresponding to each of the plurality of action commands. Here, the embedding vector corresponding to each of the plurality of action commands refers to a vector pre-allocated for each action command by the developer as a vector for expressing each action command in the form of data. The developer may assign such an embedding vector based on the degree of correlation between each action command. Specifically, the developer may assign an embedding vector corresponding to each action command such that the higher the degree of correlation between the action commands, the higher the vector similarity is calculated. For example, if there is an action command instructing to run a document creation program and an action command instructing to open a document explorer within the document creation program, the two action commands may be assigned embedding vectors such that the vector similarity is calculated high between them because they are action commands with a high degree of correlation, i.e., a high possibility that a user will execute the action commands consecutively. On the other hand, since the action command instructing to run a document creation program and the action command instructing to run a browser are action commands that have a relatively low probability of a user executing the two action commands consecutively, i.e., have a relatively low correlation, the embedding vectors may be assigned such that the vector similarity between these action commands is calculated low. In this way, the embedding vectors for each of the plurality of action commands may be assigned such that the vector similarity between action commands with a high correlation is calculated high, and the vector similarity between action commands with a low correlation is calculated low, according to the correlation between the respective action commands.

[0023] When a command to activate the action command recommendation function is authorized by a user of a computing device (110), the download unit (112) downloads data for a command recommendation model from a preset model management system server (10).

[0024] Here, the model management system server (10) refers to a server that stores and manages data on the command recommendation model required for the computing device (110) of the present invention to realize a function of recommending the next action command by analyzing the user's action command execution history.

[0025] At this time, the command recommendation model means an artificial intelligence model that has been pre-machine-learned to select the next action command corresponding to the command set input from among the plurality of action commands when a command set consisting of k (k is a natural number greater than or equal to 2) action commands sequentially executed on a computing device (110) is input as an input.

[0026] Specifically, the above command recommendation model may be a model created by training a Transformer based on a plurality of preset training command sets (each of the plurality of training command sets is a set composed by sequentially arranging k action commands selected from the plurality of action commands) and a correct answer label indicating one action command corresponding to each training command set.

[0027] Here, a transformer refers to an artificial intelligence neural network model that produces an output corresponding to a given input sequence when a given input sequence is applied. The transformer is composed of an encoder and a decoder, and is configured so that self-attention is performed in each of the encoder and decoder, and finally, attention is performed between the encoder and decoder.

[0028] At this time, the learning for the transformer may be performed so that when the embedding input for each of the plurality of training instruction sets (the embedding input for each of the plurality of training instruction sets is an embedding input generated by performing positional encoding based on the sort order of each of the k action instructions constituting each training instruction set and the embedding vector for each action instruction) is passed through the transformer to produce a prediction value, the error between the prediction value corresponding to each training instruction set and the correct answer label corresponding to each training instruction set may be minimized.

[0029] For example, let us assume that one of the plurality of training instruction sets is composed of '[action instruction 1, action instruction 2, action instruction 3]', and the correct answer label corresponding to the training instruction set is designated as a correct answer label indicating 'action instruction 5'.

[0030] Then, the learning for the transformer is performed by performing positional encoding based on the sorting order of 'action instruction 1, action instruction 2, action instruction 3' and the embedding vectors corresponding to each of 'action instruction 1, action instruction 2, action instruction 3', thereby creating an embedding input corresponding to the training instruction set, and then passing the embedding input through the transformer so that the predicted value produced is as close as possible to the correct label indicating 'action instruction 5', in other words, so that the error is minimized.

[0031] At this time, the operator who operates the model management system server (10) can periodically secure a training dataset consisting of a new training command set and its corresponding correct answer label, and then retrain the transformer based on this training dataset, thereby performing a model update so that the command recommendation model can produce more accurate results.

[0032] In this situation, when a command to activate the action command recommendation function is issued to the computing device (110) from the user of the computing device (110), the download unit (112) can download data for the command recommendation model from the model management system server (10).

[0033] In this way, when data for the above command recommendation model is downloaded to the computing device (110) of the present invention, the judgment event generation unit (113) generates a judgment event for determining whether to execute the action command recommendation function whenever the cumulative number of action commands executed by the user in the computing device (110) reaches k.

[0034] For example, when k is '3', the judgment event generation unit (113) can generate the judgment event in the computing device (110) whenever the cumulative number of action commands executed by the user reaches 3.

[0035] The execution event generating unit (114) checks whether an additional action command by the user is executed on the computing device (110) within a preset waiting time from the time when the judgment event occurs through the judgment event generating unit (113) at a certain point in time, and if it is confirmed that it is not executed, the generation of the judgment event through the judgment event generating unit (113) is stopped, and then an execution event is generated to execute the action command recommendation function.

[0036] In relation to this, since the previous judgment event has occurred, as three action commands are sequentially executed in the computing device (110), if it is determined at a certain point in time that a judgment event has occurred through the judgment event generating unit (113), the execution event generating unit (114) can check whether an additional action command by the user is executed in the computing device (110) within a preset waiting time (e.g., '10 seconds') from the time that the judgment event has occurred. If it is determined that the additional action command by the user is executed within '10 seconds', the execution event generating unit (114) does not perform any action and processes the judgment event generation process of the next cycle to be performed through the judgment event generating unit (113). However, if it is confirmed that no additional action command is executed by the user within '10 seconds', it can be seen that the user is considering what action command to input next, so the execution event generation unit (114) can process the generation of the judgment event through the judgment event generation unit (113) to be stopped, and then generate an execution event for executing the action command recommendation function.

[0037] In this way, when an execution event is generated through the execution event generation unit (114), the designation unit (115) checks k first action commands that have been cumulatively executed on the computing device (110) from the time when the previous judgment event occurred to the time when the current judgment event occurred, sorts the first action commands according to their execution order to form a first command set, and then inputs the first command set as an input to the command recommendation model, thereby processing so that one action command among the plurality of action commands is selected through the command recommendation model, and when one action command is selected through the command recommendation model, the selected action command is designated as a recommended action command.

[0038] At this time, according to one embodiment of the present invention, when the first instruction set is configured, the designation unit (115) checks the embedding vector of each of the first action instructions constituting the first instruction set by referring to the instruction database (111), and then performs positional encoding based on the sorting order of each of the first action instructions in the first instruction set and the embedding vector for each of the first action instructions, thereby generating an embedding input corresponding to the first instruction set, and then passing the generated embedding input to the transformer constituting the instruction recommendation model, thereby selecting any one of the plurality of action instructions corresponding to the first instruction set.

[0039] In relation to this, if the first action commands are 'action command A, action command B, action command C', the designation unit (115) can organize 'action command A, action command B, action command C' according to their execution order to form a first command set called '[action command A, action command B, action command C]'.

[0040] After that, the designation unit (115) can refer to the command database (111) to check the embedding vector of each of the 'action command A, action command B, and action command C' that constitute the first command set.

[0041] Then, the designation unit (115) performs positional encoding based on the sorting order in the first instruction set, which is 'action instruction A, action instruction B, action instruction C', and the embedding vectors corresponding to each of 'action instruction A, action instruction B, and action instruction C', thereby creating an embedding input corresponding to the first instruction set, and then passes the embedding input through the transformer, thereby processing so that one action instruction corresponding to the first instruction set among the plurality of action instructions is selected.

[0042] For example, if 'action command X' is selected as one of the action commands corresponding to the first command set, the designation unit (115) can designate the selected 'action command X' as a recommended action command.

[0043] In this way, when the designation of the above recommended action command is completed, the guidance unit (116) generates a guidance message guiding that the above recommended action command is recommended as the next action command corresponding to the above first action commands, displays it on the screen, and then processes the generation of the judgment event through the judgment event generation unit (113) to be resumed.

[0044] In relation to this, as in the example described above, if the recommended action command is designated as 'action command X', the guidance unit (116) can generate a guidance message suggesting whether to execute 'action command X' as a recommended action command to be executed after 'action command A, action command B, action command C' and display it on the screen. Then, the guidance unit (116) can process so that the generation of the judgment event through the judgment event generation unit (113) is resumed. Through this, the judgment event generation unit (113) can process so that the action command recommendation function continues to operate in the computing device (110) by performing an operation of generating a judgment event again when the cumulative execution number of action commands becomes 3.

[0045] According to one embodiment of the present invention, the computing device (110) may further include a training data generation unit (117) and a feedback unit (118).

[0046] After the guidance message is displayed on the screen, when a new action command is executed by the user on the computing device (110), the training data generation unit (117) compares whether the new action command is the same as the recommended action command, and if it is determined that they are not the same, it generates a training data set including the new action command as correct answer data corresponding to the first command set together with the first command set.

[0047] In this regard, let us assume that a guidance message is displayed on the screen suggesting whether to execute 'action command X', as in the example described above.

[0048] At this time, when a new action command is executed by the user on the computing device (110), the training data generation unit (117) can compare whether the new action command is the same as the recommended action command, 'action command X'.

[0049] If the new action command is determined to be not the same as the recommended action command, 'action command X', then the recommendation of the action command following 'action command A, action command B, action command C' may be considered incorrect, and therefore the training data generation unit (117) transmits feedback information about this to the model management system server (10), thereby inducing the operator of the model management system server (10) to improve the command recommendation model stored on the model management system server (10) based on this feedback information.

[0050] Specifically, the training data generation unit (117) can generate a training data set including the new action command as correct answer data corresponding to the first command set, along with the first command set, which is '[action command A, action command B, action command C]', which is composed of 'action command A, action command B, and action command C', as shown in Table 1 below.

[0051]

[0052] Command Set Answer Data [Action Command A, Action Command B, Action Command C] New Action Command

[0053]

[0054] In this way, when the training dataset is created, the feedback unit (118) transmits the training dataset to the model management system server (10).

[0055] Then, the operator who operates the model management system server (10) can update the command recommendation model stored in the model management system server (10) by retraining the transformer that constitutes the command recommendation model after specifying the command set as input in the training data set as in Table 1, specifying the correct label corresponding to the new action command specified as correct data as output, and then retraining the transformer that constitutes the command recommendation model.

[0056] At this time, according to one embodiment of the present invention, the feedback unit (118) may further include a configuration for encrypting and transmitting the training data set when transmitting the training data set to the model management system server (10) in order to strengthen data security.

[0057] In this regard, the feedback unit (118) may include a function database (119), a transmission event generation unit (120), an encryption unit (121), and a feedback processing unit (122).

[0058] The function database (119) stores a preset OTP (One-Time Password) generation function for generating a one-time password of t digits (where t is a natural number of 6 or more), a preset random number generation function, and a preset hash function. At this time, the OTP generation function, the random number generation function, and the hash function are also stored in the model management system server (10).

[0059] Here, the OTP generation function is a function that can generate the same one-time password of t digits according to the same number generation algorithm using information that can be commonly verified between the transmitting side and the receiving side (e.g., time) as a seed, and is a function frequently used in environments requiring security, such as Internet banking. In addition, the random number generation function refers to a function that generates a random number by receiving a predetermined seed as input.

[0060] When the training data set is generated through the training data generation unit (117), the transmission event generation unit (120) generates a transmission event to encrypt and transmit the training data set to the model management system server (10).

[0061] When the transmission event occurs, the encryption unit (121) generates a first one-time password of t digits through the OTP generation function, and then applies each digit of the first one-time password as a seed to the random number generation function, thereby generating t first random numbers, and then applies the first random numbers as seeds to the random number generation function again, thereby generating t second random numbers, and then sorts the first random numbers and the second random numbers in ascending order, calculates a Pearson Correlation Coefficient (PCC) between the first random numbers sorted in ascending order and the second random numbers sorted in ascending order, applies the PCC as an input to the hash function, generates a hash value, and then encrypts the training data set based on the hash value.

[0062] For example, if t is '8' and the first one-time password generated through the OTP generation function is '23512397', the encryption unit (121) can generate eight first random numbers by applying the numbers '2, 3, 5, 1, 2, 3, 9, 7', which are the digits of each digit constituting the first one-time password, as seeds to the random number generation function, and can generate eight second random numbers by applying the first random numbers as seeds to the random number generation function once again.

[0063] At this time, if the first random numbers are 'R1, R2, R3, R4, R5, R6, R7, R8' and the second random numbers are 'A1, A2, A3, A4, A5, A6, A7, A8', the encryption unit (121) can sort the first random numbers and the second random numbers in ascending order, respectively. If the first random numbers sorted in ascending order are 'R3, R2, R4, R1, R7, R5, R6, R8' and the second random numbers sorted in ascending order are 'A7, A6, A8, A4, A3, A2, A1, A5', the encryption unit (121) can sort the first random numbers sorted in ascending order, 'R3, R2, R4, R1, R7, R5, R6, R8', and the second random numbers sorted in ascending order, PCC between 'A7, A6, A8, A4, A3, A2, A1, A5' can be calculated.

[0064] Then, the encryption unit (121) can input the PCC to the hash function to generate a hash value, and then encrypt the training data set based on the hash value.

[0065] In this way, when the encryption of the training dataset is completed, the feedback processing unit (122) transmits the encrypted training dataset to the model management system server (10).

[0066] At this time, according to one embodiment of the present invention, when the encrypted training data set is received from the computing device (110), the model management system server (10) generates the first one-time password of t digits through the OTP generation function stored in the model management system server (10), and then applies each digit of the first one-time password as a seed to the random number generation function stored in the model management system server (10), thereby generating t first random numbers, and then applying the first random numbers as seeds once again to the random number generation function, thereby generating t second random numbers, and then sorting the first random numbers and the second random numbers in ascending order, calculating the PCC between the first random numbers sorted in ascending order and the second random numbers sorted in ascending order, and applying the PCC as an input to the hash function stored in the model management system server (10) to generate the hash value, and then Based on the value, the encrypted training dataset can be decrypted.

[0067] In relation to this, as in the example described above, if the first one-time password generated through the OTP generation function is '23512397', when the encrypted training dataset is received from the computing device (110) to the model management system server (10), the model management system server (10) can also generate the first one-time password '23512397' through the OTP generation function stored in the model management system server (10).

[0068] Then, the model management system server (10) can generate eight first random numbers by applying the numbers '2, 3, 5, 1, 2, 3, 9, 7', which are the numbers of each digit constituting the first one-time password, as seeds to the random number generation function stored in the model management system server (10). Since the eight first random numbers generated in this way are random numbers generated based on the same seed as in the encryption unit (121), they have the same values ​​as the first random numbers generated in the encryption unit (121). In addition, the model management system server (10) can also generate eight second random numbers by applying the first random numbers as seeds to the random number generation function once again.

[0069] Then, the model management system server (10) can sort the first random numbers and the second random numbers in ascending order, respectively, and calculate the PCC between the first random numbers sorted in ascending order and the second random numbers sorted in ascending order.

[0070] Thereafter, the model management system server (10) can input the PCC to the hash function stored in the model management system server (10) to generate the hash value, and then decrypt the encrypted training dataset based on the hash value.

[0071] FIG. 2 is a flowchart illustrating an operation method of a computing device capable of providing an artificial intelligence-based recommendation function that recommends the next action command by analyzing a user's action command execution history according to one embodiment of the present invention.

[0072] In step (S210), a command database is maintained in which a plurality of preset action commands (each of the plurality of action commands means a command for calling an action designated as executable on the computing device) are stored.

[0073] In step (S220), when a command to activate an action command recommendation function is received from a user of the computing device, data for a command recommendation model (the command recommendation model is an artificial intelligence model that has been pre-machine-learned to select a next action command corresponding to a command set received as input from among the plurality of action commands when a command set consisting of k (k is a natural number greater than or equal to 2) action commands sequentially executed on the computing device is received as input) is downloaded from a preset model management system server.

[0074] In step (S230), when data for the command recommendation model is downloaded, the computing device generates a judgment event for determining whether to execute an action command recommendation function whenever the cumulative number of action commands executed by the user reaches k.

[0075] In step (S240), when a judgment event occurs through step (S230) at some point, it is checked whether an additional action command by the user is executed on the computing device within a preset waiting time from the time the judgment event occurs, and if it is confirmed that it is not executed, the occurrence of the judgment event is stopped, and then an execution event for executing an action command recommendation function is generated.

[0076] In step (S250), when an execution event occurs through step (S240), k first action commands that have been cumulatively executed on the computing device from the time when a previous judgment event occurred to the time when the judgment event occurred are checked, the first action commands are sorted according to their execution order to form a first command set, and then the first command set is input to the command recommendation model, so that one action command among the plurality of action commands is selected through the command recommendation model. Then, when one action command is selected through the command recommendation model, the selected action command is designated as a recommended action command.

[0077] In step (S260), a guidance message is generated and displayed on the screen to guide the recommendation of the recommended action command as the next action command corresponding to the first action commands, and then the occurrence of the judgment event that was interrupted is processed to be resumed.

[0078] At this time, according to one embodiment of the present invention, the command database may additionally store preset embedding vectors corresponding to each of the plurality of action commands.

[0079] At this time, the command recommendation model is a model created by training a transformer based on a plurality of preset training command sets (each of the plurality of training command sets is a set configured by sequentially arranging k action commands selected from among the plurality of action commands) and a correct label indicating one action command corresponding to each training command set, and the training of the transformer may be performed such that when an embedding input for each of the plurality of training command sets (the embedding input for each of the plurality of training command sets is an embedding input generated by performing positional encoding based on the sorting order of each of the k action commands constituting each training command set and the embedding vector for each action command) is passed through the transformer to produce a prediction value, an error between the prediction value corresponding to each training command set and the correct label corresponding to each training command set may be minimized.

[0080] At this time, according to one embodiment of the present invention, in step (S250), when the first instruction set is configured, the instruction database is referenced to confirm the embedding vector of each of the first action instructions constituting the first instruction set, and then positional encoding is performed based on the sorting order of each of the first action instructions in the first instruction set and the embedding vector for each of the first action instructions, thereby generating an embedding input corresponding to the first instruction set, and then passing the generated embedding input to the transformer constituting the instruction recommendation model, thereby selecting one of the plurality of action instructions corresponding to the first instruction set.

[0081] In addition, according to one embodiment of the present invention, the operating method of the computing device may further include, when a new action command is executed by the user on the computing device after the guidance message is displayed on the screen, comparing whether the new action command is the same as the recommended action command, and if it is determined that the new action command is not the same, generating a training data set including the new action command as correct answer data corresponding to the first command set together with the first command set, and when the training data set is generated, transmitting the training data set to the model management system server.

[0082] At this time, according to one embodiment of the present invention, the transmitting step comprises: a step of maintaining a function database in which a preset OTP (One-Time Password) generating function for generating a one-time password of t (where t is a natural number of 6 or more), a preset random number generating function, and a preset hash function are stored (the OTP generating function, the random number generating function, and the hash function are also stored in the model management system server); a step of generating a transmission event for encrypting and transmitting the training dataset to the model management system server when the training dataset is generated; when the transmission event is generated, a first one-time password of t digits is generated through the OTP generating function, and then, by applying each number of digits constituting the first one-time password as a seed to the random number generating function, t first random numbers are generated, and by applying the first random numbers as seeds to the random number generating function once again, t second random numbers are generated, and then, the first random numbers and the second random numbers are each The method may include a step of sorting in ascending order, calculating a Pearson Correlation Coefficient (PCC) between the first random numbers sorted in ascending order and the second random numbers sorted in ascending order, applying the PCC as an input to the hash function to generate a hash value, and then encrypting the training data set based on the hash value, and transmitting the encrypted training data set to the model management system server.

[0083] At this time, when the encrypted training dataset is received from the computing device, the model management system server generates the first one-time password having t digits through the OTP generation function stored in the model management system server, and then generates t first random numbers by applying each digit constituting the first one-time password as a seed to the random number generation function stored in the model management system server, and then generates t second random numbers by applying the first random numbers as seeds again to the random number generation function, and then sorts the first random numbers and the second random numbers in ascending order, and calculates the PCC between the first random numbers sorted in ascending order and the second random numbers sorted in ascending order, and applies the PCC as an input to the hash function stored in the model management system server to generate the hash value, and then decrypts the encrypted training dataset based on the hash value.

[0084] Above, the method of operating a computing device according to an embodiment of the present invention has been described with reference to FIG. 2. Here, the method of operating a computing device according to an embodiment of the present invention may correspond to the configuration of the operation of the computing device (110) described using FIG. 1, and therefore, a more detailed description thereof will be omitted.

[0085] The operating method of a computing device according to one embodiment of the present invention can be implemented as a computer program stored in a storage medium for execution through combination with a computer.

[0086] In addition, the operating method of a computing device according to an embodiment of the present invention may be implemented in the form of program commands that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program commands, data files, data structures, etc., alone or in combination. The program commands recorded on the medium may be those specially designed and configured for the present invention or may be those known to and usable by those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and hardware devices specially configured to store and execute program commands, such as ROMs, RAMs, and flash memories. Examples of program commands include not only machine language codes generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter, etc.

[0087] As described above, the present invention has been described with specific details such as specific components and limited examples and drawings, but these are provided only to help a more general understanding of the present invention, and the present invention is not limited to the above examples, and those with ordinary knowledge in the field to which the present invention pertains can make various modifications and variations from this description.

[0088] Therefore, the idea of ​​the present invention should not be limited to the described embodiments, and all things that are equivalent or equivalent to the claims described below as well as the claims are considered to fall within the scope of the idea of ​​the present invention.

Claims

1. A computing device capable of providing an artificial intelligence-based recommendation function that recommends the next action command by analyzing the user's action command execution history, A command database storing a plurality of preset action commands, each of said plurality of action commands being a command for invoking an action designated as executable on said computing device; When a command to activate an action command recommendation function is authorized from a user of the computing device, a download unit that downloads data on a command recommendation model from a preset model management system server, wherein the command recommendation model is an artificial intelligence model that has been pre-machine-learned to select a next action command corresponding to a command set authorized as an input among the plurality of action commands when a command set consisting of k (k is a natural number greater than or equal to 2) action commands sequentially executed on the computing device among the plurality of action commands is authorized as an input; When data for the above command recommendation model is downloaded, a judgment event generation unit that generates a judgment event for determining whether to execute an action command recommendation function whenever the cumulative number of action commands executed by the user reaches k in the computing device; At a certain point in time, when a judgment event is generated through the judgment event generating unit, an execution event generating unit that checks whether an additional action command by the user is executed on the computing device within a preset waiting time from the time when the judgment event is generated, and if it is confirmed that it is not executed, processes the generation of the judgment event through the judgment event generating unit to be stopped, and then generates an execution event for executing an action command recommendation function; Through the execution event generating unit, when an execution event occurs, a designation unit that checks k first action commands that have been cumulatively executed on the computing device from the time a previous judgment event occurred to the time a current judgment event occurs, sorts the first action commands according to their execution order to form a first command set, and then inputs the first command set as an input to the command recommendation model, thereby processing the command recommendation model to select one of the plurality of action commands, and when one of the action commands is selected through the command recommendation model, designates the selected action command as a recommended action command; and A guidance unit that generates a guidance message guiding the recommendation of the above recommended action command as the next action command corresponding to the first action commands and displays it on the screen, and then processes the generation of the judgment event through the judgment event generation unit to resume. A computing device comprising:

2. In paragraph 1, The above command database additionally stores preset embedding vectors corresponding to each of the plurality of action commands, The above command recommendation model is a model created by training a transformer based on a plurality of preset training command sets, each of the plurality of training command sets being a set formed by sequentially arranging k action commands selected from among the plurality of action commands, and a correct label indicating one action command corresponding to each training command set, wherein the training of the transformer is performed such that when an embedding input for each of the plurality of training command sets, the embedding input for each of the plurality of training command sets, is an embedding input generated by performing positional encoding based on the sorting order of each of the k action commands constituting each training command set and the embedding vector for each action command, is passed through the transformer to produce a prediction value, the error between the prediction value corresponding to each training command set and the correct label corresponding to each training command set is minimized.

3. In paragraph 2, The above designated department A computing device characterized in that, when the first instruction set is configured, the instruction database is referenced to confirm the embedding vector of each of the first action instructions constituting the first instruction set, and then positional encoding is performed based on the sort order of each of the first action instructions in the first instruction set and the embedding vector for each of the first action instructions, thereby generating an embedding input corresponding to the first instruction set, and then passing the generated embedding input through the transformer constituting the instruction recommendation model, thereby selecting one of the plurality of action instructions corresponding to the first instruction set.

4. In paragraph 1, After the above guidance message is displayed on the screen, when a new action command is executed by the user on the computing device, a training data generation unit compares whether the new action command is the same as the recommended action command, and if it is determined that they are not the same, generates a training data set including the new action command as correct answer data corresponding to the first command set together with the first command set; and When the above training dataset is created, a feedback unit that transmits the training dataset to the model management system server A computing device further comprising:

5. In paragraph 4, The above feedback section A function database storing a preset OTP (One-Time Password) generation function for generating a one-time password of t (where t is a natural number greater than or equal to 6), a preset random number generation function, and a preset hash function; - The OTP generation function, the random number generation function, and the hash function are also stored in the model management system server - When the training dataset is generated, a transmission event generation unit that generates a transmission event to encrypt and transmit the training dataset to the model management system server; When the above transmission event occurs, an encryption unit that generates a first one-time password of t digits through the OTP generation function, and then applies each digit of the first one-time password as a seed to the random number generation function to generate t first random numbers, and then generates t second random numbers by applying the first random numbers as seeds to the random number generation function again, and then sorts the first random numbers and the second random numbers in ascending order, calculates a Pearson Correlation Coefficient (PCC) between the first random numbers sorted in ascending order and the second random numbers sorted in ascending order, applies the PCC as an input to the hash function to generate a hash value, and then encrypts the training data set based on the hash value; and A feedback processing unit that transmits the encrypted training dataset to the model management system server. Including, The above model management system server A computing device characterized in that, when the encrypted training dataset is received from the computing device, the first one-time password having t digits is generated through the OTP generation function stored in the model management system server, and then each digit constituting the first one-time password is applied as a seed to the random number generation function stored in the model management system server, thereby generating t first random numbers, and then the first random numbers are applied as seeds again to the random number generation function, thereby generating t second random numbers, and then the first random numbers and the second random numbers are sorted in ascending order, and the PCC between the first random numbers sorted in ascending order and the second random numbers sorted in ascending order is calculated, and the PCC is applied as an input to the hash function stored in the model management system server, thereby generating the hash value, and then the encrypted training dataset is decrypted based on the hash value.

6. A method of operating a computing device capable of providing an artificial intelligence-based recommendation function that recommends the next action command by analyzing the user's action command execution history, A step of maintaining a command database in which a plurality of preset action commands are stored, each of the plurality of action commands being a command for invoking an action designated as executable on the computing device; A step of downloading data for a command recommendation model from a preset model management system server when a command to activate an action command recommendation function is authorized from a user of the computing device, wherein the command recommendation model is an artificial intelligence model that has been pre-machine-learned to select a next action command corresponding to a command set authorized as an input among the plurality of action commands when a command set consisting of k (k is a natural number greater than or equal to 2) action commands sequentially executed on the computing device among the plurality of action commands is authorized as an input; When data for the above command recommendation model is downloaded, a step of generating a judgment event for determining whether to execute an action command recommendation function whenever the cumulative number of action commands executed by the user reaches k in the computing device; A step of generating the judgment event at a certain point in time, when the judgment event is generated, checking whether an additional action command by the user is executed on the computing device within a preset waiting time from the time the judgment event is generated, and if it is confirmed that it is not executed, processing to stop the generation of the judgment event, and then generating an execution event for executing an action command recommendation function; Through the step of generating the execution event, when the execution event occurs, a step of checking k first action commands that have been cumulatively executed on the computing device from the time when the previous judgment event occurred to the time when the judgment event occurred, arranging the first action commands in the execution order to form a first command set, and then inputting the first command set as an input to the command recommendation model, thereby processing the command recommendation model to select one of the plurality of action commands, and when one of the action commands is selected through the command recommendation model, designating the selected action command as a recommended action command; and A step of generating a guidance message that recommends the above recommended action command as the next action command corresponding to the first action commands and displaying it on the screen, and then processing the occurrence of the judgment event that was interrupted to resume. A method of operating a computing device comprising:

7. In paragraph 6, The above command database additionally stores preset embedding vectors corresponding to each of the plurality of action commands, The above command recommendation model is a model created by training a transformer based on a plurality of preset training command sets, each of the plurality of training command sets being a set formed by sequentially arranging k action commands selected from among the plurality of action commands, and a correct label indicating one action command corresponding to each training command set, wherein the training of the transformer is performed such that when an embedding input for each of the plurality of training command sets, the embedding input for each of the plurality of training command sets, is an embedding input generated by performing positional encoding based on the sorting order of each of the k action commands constituting each training command set and the embedding vector for each action command, is passed through the transformer to produce a prediction value, an error between the prediction value corresponding to each training command set and the correct label corresponding to each training command set is minimized.

8. In paragraph 7, The steps specified above are A method of operating a computing device, characterized in that when the first instruction set is configured, the instruction database is referenced to confirm the embedding vector of each of the first action instructions constituting the first instruction set, and then positional encoding is performed based on the sort order of each of the first action instructions in the first instruction set and the embedding vector for each of the first action instructions, thereby generating an embedding input corresponding to the first instruction set, and then passing the generated embedding input through the transformer constituting the instruction recommendation model, thereby processing so that one of the plurality of action instructions corresponding to the first instruction set is selected.

9. In paragraph 6, After the above guidance message is displayed on the screen, when a new action command is executed by the user on the computing device, comparing whether the new action command is the same as the recommended action command, and if it is determined that they are not the same, generating a training data set including the new action command as correct answer data corresponding to the first command set together with the first command set; and When the above training dataset is created, a step of transmitting the training dataset to the model management system server A method of operating a computing device further comprising:

10. In paragraph 9, The above transmitting steps are A step of maintaining a function database in which a preset OTP (One-Time Password) generation function for generating a one-time password of t (where t is a natural number greater than or equal to 6), a preset random number generation function, and a preset hash function are stored; - The OTP generation function, the random number generation function, and the hash function are also stored in the model management system server - When the training dataset is generated, a step of generating a transmission event for encrypting and transmitting the training dataset to the model management system server; When the above transmission event occurs, a step of generating a first one-time password of t digits through the OTP generation function, and then applying each digit constituting the first one-time password as a seed to the random number generation function to generate t first random numbers, and then generating t second random numbers by applying the first random numbers as seeds to the random number generation function again, and then sorting the first random numbers and the second random numbers in ascending order, calculating a Pearson Correlation Coefficient (PCC) between the first random numbers sorted in ascending order and the second random numbers sorted in ascending order, and applying the PCC as an input to the hash function to generate a hash value, and then encrypting the training data set based on the hash value; and A step of transmitting the encrypted training dataset to the model management system server. Including, The above model management system server A method of operating a computing device, characterized in that when the encrypted training dataset is received from the computing device, the first one-time password having t digits is generated through the OTP generation function stored in the model management system server, and then each digit constituting the first one-time password is applied as a seed to the random number generation function stored in the model management system server, thereby generating t first random numbers, and then the first random numbers are applied as seeds again to the random number generation function, thereby generating t second random numbers, and then the first random numbers and the second random numbers are sorted in ascending order, and the PCC between the first random numbers sorted in ascending order and the second random numbers sorted in ascending order is calculated, and the PCC is applied as an input to the hash function stored in the model management system server, thereby generating the hash value, and then the encrypted training dataset is decrypted based on the hash value.

11. A computer-readable recording medium recording a computer program for executing the method of any one of clauses 6 to 10 through combination with a computer.

12. A computer program stored in a storage medium for executing the method of any one of clauses 6 to 10 through combination with a computer.

Images