KR101988028B1 - Server and operating method of the same - Google Patents

Abstract

A processing method is disclosed. According to an embodiment of the present invention, the processing method comprises the steps of: receiving an input vector representing time series data as a vector; compressing at least one feature of the received input vector by encoding based on the received input vector; and drawing result information by decoding based on the compressed feature.

Description

[0001] DESCRIPTION [0002] SERVER AND OPERATING METHOD OF THE SAME [0003]

The following embodiments relate to data processing.

Convolutional Neural Network (CNN) is a neural network that uses convolution operations.

Related Prior Art US Patent Publication No. US 2018/0075343 (entitled PROCESSING SEQUENCES USING CONVOLUTIONAL NEURAL NETWORKS, filed by Google) is available. The neural network system disclosed in the patent publication generates an output sequence of audio data. Here, the neural network system includes a convolutional subnetwork, which includes a CNN layer and an output layer for processing one or more audio data.

A method according to one aspect of the present invention includes receiving an input vector representing a time series data as a vector; Performing an encoding based on the received input vector to compress at least one feature of the received input vector; And performing decoding based on the compressed characteristic to derive result information.

The processing method may further include calculating a recall rate indicating how much the derived result information reproduces the received input vector.

The processing method may further include detecting clustering type change through a type of clustering based on the compressed characteristic and a type of previous clustering.

The time series data may include game log data.

The encoding may be performed by an encoder in a convolutional neural networks (CNN) based auto-encoder, and the decoding may be performed by a decoder in the auto-encoder.

The processing method according to another aspect includes: generating an input vector from time series data using a vector generator; And deriving result information corresponding to the generated input vector using a processor.

The processor performs encoding based on the generated input vector, compresses at least one feature of the generated input vector, and performs decoding based on the compressed feature to derive the result information.

The processing method may further include calculating a recall rate indicating how much the derived result information reproduces the received input vector.

The processing method may further include detecting clustering type change through a type of clustering based on the compressed characteristic and a type of previous clustering.

The time series data may include game log data.

The processor may correspond to a CNN based auto encoder.

A processing device according to one of the claims, wherein the processing device is configured to process the time series data to receive the generated input vector, obtain at least one feature from the received input vector, compress the obtained feature, And derives result information by decoding.

The controller can calculate a recall rate indicating how much the derived result information reproduces the received input vector.

The controller may detect a clustering form change through a type of clustering based on the compressed characteristics and a type of previous clustering.

The time series data may include game log data.

The processing unit according to one side includes a memory for storing a CNN (Convolutional Neural Networks) based processing model; And an input vector generated from the time series data and a controller for deriving the result information using the CNN based processing model.

The controller performs encoding based on the generated input vector, compresses at least one feature of the generated input vector, and performs decoding based on the compressed feature to derive the result information.

The controller can calculate a recall rate indicating how much the derived result information reproduces the received input vector.

The controller may detect a clustering form change through a type of clustering based on the compressed characteristics and a type of previous clustering.

The time series data may include game log data.

The CNN-based processing model may correspond to a CNN-based auto encoder.

Embodiments can detect sparse data, estimate the diversity of input data during learning, and process multidimensional data. In other words, embodiments can observe trends in sparse data, observe trends in data diversity, and compress characteristics of multidimensional data to track clustering form changes. Accordingly, embodiments can build a versatile and scalable control system (e.g., a system for control of game logs, server logs, security logs, etc.) and efficiently process multi-dimensional data for control .

1 is a view for explaining a game system according to an embodiment.
FIGS. 2 to 5 are diagrams for explaining an example of the operation of the vector generator according to one embodiment.
6 is a diagram for explaining a learning process according to an embodiment.
7 to 8 are views for explaining a processor according to an embodiment.
9 is a block diagram for explaining a processing apparatus according to an embodiment.
10 is a flowchart for explaining an example of a processing method according to an embodiment.
11 is a flowchart for explaining another example of a processing method according to an embodiment.

In the following, embodiments will be described in detail with reference to the accompanying drawings. However, various modifications may be made in the embodiments, and the scope of the patent application is not limited or limited by these embodiments. It is to be understood that all changes, equivalents, and alternatives to the embodiments are included in the scope of the right.

The terms used in the examples are used for descriptive purposes only and are not to be construed as limiting. The singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, the terms "comprises" or "having" and the like refer to the presence of stated features, integers, steps, operations, elements, components, or combinations thereof, But do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or combinations thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this embodiment belongs. Terms such as those defined in commonly used dictionaries are to be interpreted as having a meaning consistent with the contextual meaning of the related art and are to be interpreted as either ideal or overly formal in the sense of the present application Do not.

In the following description of the present invention with reference to the accompanying drawings, the same components are denoted by the same reference numerals regardless of the reference numerals, and redundant explanations thereof will be omitted. In the following description of the embodiments, a detailed description of related arts will be omitted if it is determined that the gist of the embodiments may be unnecessarily blurred.

1 is a view for explaining a game system according to an embodiment.

Referring to FIG. 1, a game system 100 according to an embodiment includes a server 110, a terminal 120, and a database 130.

The terminal 120 may correspond to a mobile terminal such as a smart phone, a tablet, and / or a fixed terminal such as a PC.

The terminal 120 may be provided with a game client. The terminal 120 can access the server 110 through the corresponding game client. The game client may be represented differently as a game application or game software.

One terminal 120 is illustrated in FIG. 1, but this is only an example, and a plurality of terminals can connect to the server 110.

The server 110 according to one embodiment includes a vector generator 111 and a processor 112.

The vector generator 111 acquires a plurality of data (or game data) including time information. The data (or game data) may be stored, for example, in the database 130, and the vector generator 111 may obtain the corresponding data (or game data) from the database 130 .

The vector generator 111 generates an input vector based on the obtained data. Where the axis of the input vector represents the time axis and the other axis of the input vector represents the axis based on at least one user defined field. The input vector corresponds to a multidimensional vector. In other words, the input vector corresponds to a vector of two or more dimensions. A method by which the vector generator 111 generates an input vector will be described later with reference to FIG. 2 to FIG.

The vector generator 111 inputs or transmits the generated input vector to the processor 112.

The processor 112 performs a convolution operation based on the input vector. In other words, the processor 112 includes convolutional neural networks (CNN), and can perform a convolution operation on the basis of the input vector received from the vector generator 111.

The output result of the processor 112 may be used for game control.

One embodiment may transform the time series data into a multidimensional vector suitable for CNN. In other words, one embodiment can lead to a new way of processing time series data. Also, one embodiment can process data that is difficult to process with an RNN (Recurrent Neural Network).

FIGS. 2 to 5 are diagrams for explaining an example of the operation of the vector generator according to one embodiment.

Referring to Fig. 2, the vector generator 111 acquires the data 210-1 to 210-n. Here, each of the data 210-1 to 210-n includes time information (or time field). In one example, the vector generator 111 may obtain the data 210-1 through 210-n from the database 130. [

The vector generator 111 generates an input vector 220 based on the data 210-1 to 210-n. As in the example shown in FIG. 2, one axis of the input vector 220 represents the time axis and the other axis represents the field axis. Here, the field axis corresponds to an axis based on at least one user defined field.

The vector generator 111 sends the input vector 220 to the processor 112.

The processor 112 performs various operations based on the input vector 220 to derive the result information. The derived result information can be used, for example, for game control. The processor 112 will be described later.

Detailed operation of the vector generator 111 will be described with reference to FIG.

Referring to Fig. 3, the vector generator 111 may obtain the data 310-1 to 310-n. The data 310-1 to 310-n are examples of the data 210-1 to 210-n, and include various fields as well as a time field. More specifically, the data 310-1 may include, for example, a time field = t-1, a field A = a-1, a field B = b-1, a field C = c-1, Field A may indicate, for example, a field relating to game connection, field B may indicate, for example, a field relating to login failure, field C may indicate, for example, . Data 310-n includes, for example, time field = t-n, field A = a-n, field B = b-n, field D = d-n, The field D may represent, for example, a field relating to a game item.

The vector generator 111 may generate an input vector 320 based on the number of data including information corresponding to time intervals and user defined fields. For example, referring to the example shown in FIG. 3, the vector generator 111 includes information corresponding to a user defined field for each time interval (0 to? 1,? 1 to? 2,? 2 to? 3,? 3 to? It is possible to count the number of pieces of data. In other words, in the example shown in FIG. 3, if it is assumed that each of the user-defined fields constituting the field axis correspond to field A, field B, and field D, the vector generator 111 calculates , How many data having field A, how many data having field B, and how many data having field D can be obtained. In this manner, the vector generator 111 can determine, based on each time period, how many data having the field A, how many data having the field B, and how many data having the field D . In the example shown in FIG. 3, it is assumed that the time field t-1 of the data 310-1 belongs to the time interval 0 to? 1. In this case, the vector generator 111 generates an element 320-1 corresponding to the time interval 0 to? 1 and the field A, a time interval 0 to? 1 and an element 320-2 corresponding to the field B in the input vector 320, The data 310-1 may be counted or recorded. It is also assumed that the time field t-n of the data 310-n belongs to the time period? 3 to? 4. In this case, the vector generator 111 generates a time interval τ3 to τ4, an element 320-3 corresponding to the field A, a time interval τ3 to τ4 and an element 320-4 corresponding to the field D in the input vector 320, May count or record that data 310-n belongs to. In this manner, the vector generator 111 may convert the data 310-1 through 310-n into an input vector 320 suitable for the input layer of the CNN.

4 is another example of the data 210-1 to 210-n. In the example shown in Fig. 4, each row represents individual data of the data 210-1 through 210-n.

In the example shown in Fig. 4, "regdatetime" represents a time field, "logid" represents a field relating to a log identifier, "adid" . "Battlepower" indicates a field related to the combat power of the game character, "playtime" indicates a field related to the game play time, "herolevel" indicates a field related to the hero level in the game, "market" indicates a game client Quot; city "indicates a field related to the city where the game connection occurred.

The vector generator 111 may convert the data shown in FIG. 4 into an input vector 220 suitable for CNN processing. At this time, the vector generator 111 can generate the input vector 220 only for a specific pid. In other words, the vector generator 111 may generate an input vector corresponding to a particular pid so that the processor 112 can perform analysis only on a specific pid. An example of a corresponding input vector is shown in Fig.

FIG. 5 shows an input vector 500 corresponding to pid # 1.

The vector generator 111 may convert the time field of each of the data of pid # 1 into the specific format among the data shown in FIG. As an example, the vector generator 111 may convert the time field of each of the data of pid # 1 into a specific scaling of "hh / mm ". In other words, the vector generator 111 can convert the "regdatetime" of each of the data of pid # 1 to "hh / mm ". Here, "hh" corresponds to hour (hour), and "mm" corresponds to minute.

The vector generator 111 may perform feature engineering on a field corresponding to a user defined field among a plurality of fields corresponding to the converted time field. For the example shown in Fig. 5, let us say that the user defined fields are "herelevel "," battlepower ", and "playtime ". The vector generator 111 can perform feature engineering on each of the fields "herelevel", "battlepower", and "playtime" among a plurality of fields corresponding to the converted time field, or aggregation. In one example, when the transformed time field 0/0 510 corresponds to "herelevel", "battlepower", and "playtime", the vector generator 111 generates "herelevel", "battlepower", and "playtime" And generate respective statistical information (e.g., standard deviation). "Herelevel", "battlepower", and "playtime" statistical information corresponding to the converted time field 0/0 (510) corresponds to "herelevel_std -0.373", "battlepower_std -0.479", and "playtime_std -0.703" The vector generator 111 may aggregate "herelevel_std -0.373 "," battlepower_std -0.479 ", and "playtime_std -0.703 " in the converted time field 0/0 (510). In this manner, the vector generator 111 may generate the input vector 500 by aggregating the statistical information of the field corresponding to the user-defined field in each of the converted time fields.

The processor 112 performs various operations based on the input vectors 220, 320, or 500 to derive the resulting information. This processor 112 can be learned in advance, and the operation of the post-processor 112 that describes the learning process will be described.

6 is a diagram for explaining a learning process according to an embodiment.

Referring to FIG. 6, a CNN based auto encoder 600 is shown. The CNN based auto encoder 600 includes an encoder and a decoder.

An encoder may be represented differently by a convolution network, and a decoder may be represented by a deconvolution network.

The encoder can extract at least one feature from the input data (or learning data) y and can compress the extracted feature. The input data y may correspond to the entire data collected by the server 110, for example.

를 생성할 수 있다. The decoder decodes at least one feature compressed by the encoder to produce output data

Lt; / RTI >

와 y 사이의 차이를 기초로 CNN 기반 오토인코더(600) 내의 가중치들을 업데이트한다. 해당 가중치들이 업데이트됨으로써 CNN 기반 오토인코더(600)는 학습된다. 달리 표현하면, CNN 기반 오토인코더(600)는 최적화 또는 피팅(fitting)된다.The CNN-based auto encoder 600

Lt; RTI ID = 0.0 > CN < / RTI > based auto encoder 600 based on the difference between y and y. The CNN-based auto-encoder 600 is learned by updating the weights. In other words, the CNN-based auto encoder 600 is optimized or fitted.

y와 y 사이의 차이는 코스트(cost) 또는 학습 코스트에 해당할 수 있다. 이러한 코스트는 입력 데이터의 다양성을 파악하는데 도움을 줄 수 있다. 다시 말해, 학습 시 최종 수렴되는 cost를 기준으로 입력 데이터의 다양성이 파악될 수 있다. 일례로, CNN 기반 오토인코더(600)는 입력 데이터 y_1을 입력 받아 출력 데이터

_1를 생성할 수 있다. 여기서,

_1과 y_1 사이의 차이가 6이라 하자. 학습 과정에서 CNN 기반 오토인코더(600)는 y_1과 동일한 사이즈의 y_2를 입력 받아 출력 데이터

_2를 생성할 수 있다. 여기서,

_2와 y_2 사이의 차이가 450이면,

_2와 y_2 사이의 차이는

_1과 y_1 사이의 차이보다 크다. 이는, 입력 데이터 y_2에는 다양한 데이터가 존재한다고 파악될 수 있다. 달리 표현하면, 과거에 수집된 데이터 y_1에 비해 최근에 수집된 y_2는 다양한 데이터를 포함할 수 있다. In an embodiment,

The difference between y and y may correspond to a cost or a learning cost. These costs can help to identify the variety of input data. In other words, the diversity of the input data can be grasped based on the final convergence cost during learning. For example, the CNN-based auto encoder 600 receives input data y_1,

_1. ≪ / RTI > here,

Let the difference between _1 and y_1 be 6. In the learning process, the CNN-based auto encoder 600 receives y_2 of the same size as y_1,

_2. ≪ / RTI > here,

If the difference between _2 and y_2 is 450,

The difference between _2 and y_2 is

Is greater than the difference between _1 and y_1. It can be understood that various data exist in the input data y_2. In other words, y_2 collected recently compared to data y_1 collected in the past may contain various data.

The learned CNN-based auto-encoder 600 may be implemented or implemented in the processor 112. The learned CNN-based auto-encoder 600 in the processor 112 can be learned every predetermined period. For example, learning can be performed daily, weekly, or monthly.

7 to 8 are views for explaining a processor according to an embodiment.

Referring to FIG. 7, a processor 112 according to one embodiment includes an encoder 710 and a decoder 720.

The encoder 710 receives the input vector 220, 320, or 500 representing the time series data as a vector. The time series data is data including time fields, for example, game log data. Since the method of expressing the time series data as a vector has been described above, a detailed description will be omitted.

The encoder 710 performs encoding based on the input vector 220, 320, or 500 to compress at least one feature of the input vector 220, 320, or 500.

The decoder 720 performs decoding based on the at least one compressed feature to derive the resulting information. The decoder 720 may compute the recall rate based on the input vector 220, 320, or 500 and the result information. The recall rate indicates how much the result information reproduces the input vector 220, 320, or 500, e.g., based on the difference between the input vector 220, 320, or 500 and the resulting information. If the processor 112 is to process the seen data, the recall rate will be calculated to be high, and if the unseen data is to be processed, the recall rate will be calculated to be low.

The processor 112 may detect sparse data through the recall ratio. In other words, the processor 112 can observe the trend of sparse data through the recall rate. In addition, the processor 112 may suspect abnormal situations through sparse data detection or sparse data trend observations. For example, if the recall rate is below the threshold, the processor 112 may suspect that an abnormal situation other than usual has occurred. Sparse data may be present in the time series data or in the input vector 220, 320, or 500, where the sparse data corresponds to unseen data that the processor 112 has not seen since the feature was not learned during the compression process. Accordingly, a difference may occur between the input and output (i. E., The input vector 220, 320, or 500) and the resulting information, and eventually the recall rate may be calculated to be low. If the recall rate is higher than the threshold value, the processor 112 may determine that there is no abnormality in the time series data or the input vector 220, 320, or 500.

In an embodiment, the processor 112 may control the cluster configuration aspect. In other words, the processor 112 can track changes in clustering form. This will be described with reference to FIG.

A CNN-based auto-encoder implemented in processor 112 is shown in FIG.

In the example shown in Fig. 8, the encoder 710 can extract and compress features A to C from the input vector. The processor 112 may perform clustering based on the compressed features A through C to generate clusters 810 through 812. The processor 112 may suspect that an abnormal situation has occurred through the previous clusters 820 to 823 and clusters 810 to 812. [ In other words, the processor 112 may suspect that the current anomaly has occurred since the currently compressed features are different in appearance from the previously compressed features. If the processor 112 generates the same clusters as the previous clusters 820 through 823 while processing the input vector, it can determine that the current abnormal situation has not occurred.

9 is a block diagram for explaining a processing apparatus according to an embodiment.

9, a processing apparatus 900 according to an embodiment includes a memory 910 and a controller 920. [

The processing device 900 corresponds to the processor 112.

The memory 910 may store a CNN-based processing model. In other words, the memory 910 may store optimized weights in the CNN-based processing model. The CNN-based processing model may correspond to the CNN-based auto encoder described above. The weights stored in the memory 910 can be optimized or learned every predetermined period through the learning process described with reference to FIG.

The controller 920 derives the result information using the input vector generated from the time-series data and the CNN-based processing model. At this time, the controller 920 performs encoding based on the input vector, compresses at least one feature of the input vector, and performs decoding based on the compressed feature to derive the result information.

The controller 920 can calculate the recall rate indicating how much the result information reproduces the input vector. The controller 920 may detect (or detect) the sparse data through the calculated recall rate to suspect that an abnormal situation has occurred.

In addition, the controller 920 can detect clustering form changes through the type of clustering based on the compressed characteristics and the type of previous clustering. Since this has been described with reference to FIG. 8, a detailed description will be omitted.

1 through 8 can be applied to the matters described with reference to FIG. 9, so that a detailed description will be omitted.

10 is a flowchart for explaining an example of a processing method according to an embodiment.

The processing method to be described with reference to FIG. 10 is performed by the processing apparatus 900 or the processor 112.

Referring to FIG. 10, the processing apparatus 900 (or the processor 112) receives an input vector representing a time series data as a vector (1010).

The processing unit 900 (or processor 112) performs an encoding based on the received input vector to compress 1020 the at least one feature of the received input vector.

The processing unit 900 (or the processor 112) performs decoding based on the compressed characteristic to derive the result information (1030).

The processing unit 900 (or the processor 112) can calculate the recall rate based on the derived result information and the input vector and calculate the recall rate based on the recall rate and the game control (e.g., game log, server log, Etc.) can be performed.

1 through 9 can be applied to the matters described with reference to FIG. 10, so that a detailed description thereof will be omitted.

11 is a flowchart for explaining another example of a processing method according to an embodiment.

The processing method to be described with reference to FIG. 11 is performed by the server 110. FIG.

Referring to FIG. 11, the server 110 generates an input vector from time series data using a vector generator 111 (1110).

The server 110 derives result information corresponding to the input vector using the processor 112 (1120). At this time, the processor 112 performs encoding based on the input vector, compresses at least one feature of the input vector, and performs decoding based on the compressed feature to derive the result information.

The server 110 may calculate the recall rate based on the derived result information and the input vector. In addition, the server 110 may perform game control (for example, monitoring a game log, a server log, a security log, and the like) based on the calculated recall rate.

1 through 9 can be applied to the matters described with reference to FIG. 10, so that a detailed description thereof will be omitted.

The apparatus described above may be implemented as a hardware component, a software component, and / or a combination of hardware components and software components. For example, the apparatus and components described in the embodiments may be implemented within a computer system, such as, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA) , A programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications running on the operating system. The processing device may also access, store, manipulate, process, and generate data in response to execution of the software. For ease of understanding, the processing apparatus may be described as being used singly, but those skilled in the art will recognize that the processing apparatus may have a plurality of processing elements and / As shown in FIG. For example, the processing unit may comprise a plurality of processors or one processor and one controller. Other processing configurations are also possible, such as a parallel processor.

The software may include a computer program, code, instructions, or a combination of one or more of the foregoing, and may be configured to configure the processing device to operate as desired or to process it collectively or collectively Device can be commanded. The software and / or data may be in the form of any type of machine, component, physical device, virtual equipment, computer storage media, or device , Or may be permanently or temporarily embodied in a transmitted signal wave. The software may be distributed over a networked computer system and stored or executed in a distributed manner. The software and data may be stored on one or more computer readable recording media.

The method according to an embodiment may be implemented in the form of a program command that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions to be recorded on the medium may be those specially designed and configured for the embodiments or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Magneto-optical media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments. For example, it is to be understood that the techniques described may be performed in a different order than the described methods, and / or that components of the described systems, structures, devices, circuits, Lt; / RTI > or equivalents, even if it is replaced or replaced.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

Claims (19)

Converting time information contained in each of the time series data;
Calculating statistical values of log information corresponding to user defined fields among game related log information included in each of the time series data; And
Generating an input vector based on the transformed time information and the calculated statistical value;
Performing an encoding based on the input vector to compress at least one feature of the input vector; And
Performing decoding based on the compressed characteristic to derive result information
/ RTI >
How the server works.
The method according to claim 1,
Calculating a recall rate indicating how much the input vector is reproduced by the derived result information
≪ / RTI >
How the server works.
The method according to claim 1,
Detecting clustering form changes through the type of clustering based on the compressed characteristics and the type of previous clustering
≪ / RTI >
How the server works.
delete The method according to claim 1,
Wherein the encoding is performed by an encoder in a convolutional neural networks (CNN) based auto-encoder, the decoding being performed by a decoder in the auto-
How the server works.
delete delete delete delete delete The time information included in each of the time series data, calculates statistical values of log information corresponding to user defined fields among the game related log information included in each of the time series data, And a controller for generating an input vector based on the calculated statistic value, obtaining at least one feature from the input vector, compressing the obtained feature, and decoding the compressed feature to derive result information.
/ RTI >
server.
12. The method of claim 11,
Wherein the controller calculates a recall rate indicating how much the input vector is reproduced by the derived result information,
server.
12. The method of claim 11,
Wherein the controller is configured to detect clustering form changes through a type of clustering based on the compressed characteristics and a type of previous clustering,
server.
delete delete delete delete delete delete

Images