KR20230076931A - Apparatus and method for detecting abnormal pattern of sound source using neural network - Google Patents

Abstract

An objective of the present invention is to promote the healthy development of a sound source market by classifying abnormal patterns. According to one embodiment of the present invention, an abnormality detection method for a sound source based on sound source information is performed by a server and comprises: (a) a step of receiving sound source information corresponding to a plurality of sound sources from a sound source server; (b) a step of inputting the sound source information of the plurality of sound sources into a preset classification model to classify types of the sound sources to generate a plurality of sound source groups; and (c) a step of inputting a specific sound source into a preset abnormality detection model corresponding to the types of the sound source groups among a plurality of abnormality detection models previously generated by learning the plurality of sound source groups to detect whether an abnormal pattern is present in the specific sound source. The sound source information consists of log data and metadata of the sound sources.

Description

Apparatus and method for detecting abnormal patterns of sound sources using artificial neural networks

The present invention is to detect a situation in which a sound source is abnormally sold or streamed using artificial neural network technology.

Recently, as the music industry develops rapidly, people's consumption trends for sound sources are also changing. Specifically, as for the method of consumption of sound sources, the streaming method is becoming the main focus in the conventional method of purchasing and downloading music due to the development of digital technology and the development of smart phones and wireless Internet. This change can be a disprove that the way of consuming music is changing from ownership to access.

These changes are accompanied by side effects that did not exist in the past. For example, artificial manipulations such as abnormal hoarding of sound sources that affect chart selection and artificial and repetitive playback of sound sources through machines (or macro techniques) may be included in order to increase the chart ranking of sound sources.

Conventionally, in order to prevent such side effects, a technique of analyzing log data on the usage flow of a sound source has been used. For example, a method of excluding sound source metadata (singer, album, issue level, etc.) through methods such as regression analysis and time series cluster analysis and statistically analyzing and classifying them, or similarity (or dissimilarity) between cluster analysis data There is a way to use . However, despite these efforts, it is difficult to fully grasp the physical flow of use.

In order to solve the above problems, the present invention clusters each sound source based on information included in the sound source using artificial neural network technology, and determines the abnormal usage flow of the sound source according to whether a specific sound source is included in the clustered sound source. The purpose is to build a system to do this.

In the anomaly detection method for a sound source based on sound source information, performed by a server according to an embodiment of the present invention for achieving the above technical problem, (a) corresponding to a plurality of sound sources from a sound source server, respectively Receiving sound source information to; (b) generating a plurality of sound source groups by inputting sound source information of a plurality of sound sources into a preset classification model and classifying the type of each sound source; and (c) detecting whether a specific sound source has an abnormal pattern by inputting a specific sound source to a preset anomaly detection model corresponding to the type of the sound source group among a plurality of anomaly detection models pre-generated by learning a plurality of sound source groups, respectively. step; Including, the sound source information may be composed of metadata and log data of the sound source.

In addition, the metadata includes at least one of the title, singer, genre, composer, lyricist, producing, release date, and featuring information of the sound source, and the log data includes information on the sales method of the sound source, information on the payment method, and whether the sound source was streamed or sold. A waveform obtained by calculating meta data and log data according to a predetermined lapse of time, including at least one of time and the number of users who have streamed or purchased the sound source, may be used as the sound source information.

In step (b), at least one or more sound source groups may be generated by applying any one of pattern-specific time-series clustering or clustering techniques to changes in sound source information.

In addition, each classified sound source group may further include labeling a type based on information having the highest similarity among sound source information included in the sound source group.

In addition, in step (c), when a specific sound source is input to the anomaly detection model that has been input by inputting a sound source group with a designated type as an input value, the type of the specific sound source can be classified together based on the type learned by the anomaly detection model. .

In addition, after step (c), if the sound source information of the specific sound source is input to the anomaly detection model and the specific sound source is output as an abnormal sound source, (d) determining that the specific sound source is in an abnormal streaming or abnormal sales state; may be further included. there is.

In addition, the anomaly detection model compares the pattern for the streaming or sales time or number of sound sources corresponding to the corresponding type, and when a different pattern is derived, the corresponding sound source may be output as an abnormal sound source group.

In addition, in step (d), the number of mentions of a specific sound source is received from a preset external server including a preset SNS and web page, and even if the anomaly detection model outputs it as an abnormal sound source group, the number of mentions exceeds the preset threshold. , the specific sound source may be determined to be in any one of normal streaming or normal sales.

Also, the number of mentions may be crawled and collected from an external server.

In addition, if the status of a specific sound source changes to a normal streaming or normal sales status after step (d), a sound source group may be newly created based on the number of mentions on SNS, and an anomaly detection model may be updated.

In addition, in the anomaly detection device for a sound source based on sound source information, a memory storing an anomaly detection program for a sound source based on the sound source information and an anomaly detection processor for a sound source based on the sound source information by executing the program stored in the memory However, the processor receives sound source information corresponding to a plurality of sound sources from the sound source server, inputs the sound source information of the plurality of sound sources to a preset classification model, classifies the type of each sound source, and generates a plurality of sound source groups. , A specific sound source is input to a preset anomaly detection model corresponding to the type of a sound source group among a plurality of anomaly detection models pre-generated by learning a plurality of sound source groups to detect whether or not there is an abnormal pattern in a specific sound source, and to detect sound source information may be a device composed of metadata and log data of a sound source.

It may be a computer readable storage medium on which a program for performing an anomaly detection method for a sound source based on the sound source information according to claim 1 is recorded.

According to an embodiment of the present invention, each sound source is clustered based on information included in the sound source using artificial neural network technology, and an abnormal usage flow of the sound source is determined according to whether a specific sound source is included in the clustered sound source system can be built.

Through this, service providers selling (streaming) music sources can accurately analyze data related to music use flows and classify abnormal patterns to promote healthy development of the music market.

1 is a diagram illustrating an anomaly detection system for a sound source based on sound source information according to an embodiment of the present invention.
2 is a diagram showing the configuration of a server according to an embodiment of the present invention.
3 is an operational flowchart illustrating a method for detecting an anomaly in a sound source based on sound source information according to an embodiment of the present invention.
4 is an operation flowchart for determining a state of a sound source through an anomaly detection model according to an embodiment of the present invention.
5 is a diagram illustrating an example of a pattern and clustering of sound source information according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail so that those skilled in the art can easily practice the present invention with reference to the accompanying drawings. However, the present invention may be embodied in many different forms and is not limited to the embodiments described herein. And in order to clearly explain the present invention in the drawings, parts irrelevant to the description are omitted, and similar reference numerals are attached to similar parts throughout the specification.

Throughout the specification, when a part is said to be "connected" to another part, this includes not only the case where it is "directly connected" but also the case where it is "electrically connected" with another element interposed therebetween. . In addition, when a part "includes" a certain component, this means that it may further include other components, not excluding other components, unless otherwise stated, and one or more other characteristics. However, it should be understood that it does not preclude the possibility of existence or addition of numbers, steps, operations, components, parts, or combinations thereof.

The following examples are detailed descriptions for better understanding of the present invention, and do not limit the scope of the present invention. Therefore, inventions of the same scope that perform the same functions as the present invention will also fall within the scope of the present invention.

1 is a diagram illustrating an anomaly detection system for a sound source based on sound source information according to an embodiment of the present invention.

Referring to FIG. 1 , the system is composed of a server 100 and a sound source server 200 . At this time, each device may be interconnected wired or wirelessly through a communication network.

According to an embodiment of the present invention, the server 100 refers to a device for generating a classification model for classifying a sound source based on sound source information. In addition, in the case of the sound source server 200, it means an external server that streams or sells a singer's sound source to people.

At this time, as an optional embodiment, the server 100 and the sound source server 200 may be combined to form one server 100, or the sound source server 200 may correspond to a terminal used by users.

According to an embodiment of the present invention, the server 100 receives sound source information on a sound source provided from the sound source server 200 and generates at least one sound source group based on this.

At this time, the classified sound source group clusters sound sources composed of information having a high degree of similarity according to information (characteristics) included in the sound source. For example, when a specific sound source (or songs of a singer who sang the corresponding sound source) consistently records steady sellers, a steady seller type sound source group is created to cluster the sound sources with steady sellers. Through this method, the server 100 generates various sound source groups according to sound source information such as one hit wonder type, popular type at dawn, type sold mainly in the evening, and type sold frequently in rainy weather through clustering of sound sources. will do

Next, the server 100 inputs a specific sound source to a plurality of anomaly detection models to detect whether or not there is an abnormal pattern in the specific sound source.

Specifically, the server 100 establishes a model for classifying the sound source according to each state by learning the learning model for the plurality of sound source groups calculated in the previous process. For example, by clustering and learning only sound sources having abnormal streaming or abnormal sales status information among a plurality of sound sources, the anomaly detection model can classify sound sources having abnormal conditions.

On the other hand, the communication network refers to a communication network that provides an access path so that data can be transmitted and received between the server 100 and the sound source server 200. Communication networks include, for example, wired networks such as LANs (Local Area Networks), WANs (Wide Area Networks), MANs (Metropolitan Area Networks), and ISDNs (Integrated Service Digital Networks), wireless LANs, wireless networks such as CDMA, Bluetooth, and satellite communication. However, the scope of the present invention is not limited thereto.

2 is a diagram showing the configuration of a server according to an embodiment of the present invention.

Referring to FIG. 2 , the server 100 according to an embodiment of the present invention includes a communication module 110, a memory 120, a processor 130, a database 140, a classification module 150, and an anomaly determination module ( 170).

In detail, the communication module 110 provides a communication interface necessary to provide a transmission/reception signal between the server 100 and the sound source server 200 in the form of packet data in conjunction with a communication network. Furthermore, the communication module 110 may serve to receive a data request from the sound source server 200 and transmit data as a response thereto.

Here, the communication module 110 may be a device including hardware and software necessary for transmitting and receiving a signal such as a control signal or a data signal with another network device through a wired or wireless connection.

The memory 120 stores a program for detecting abnormalities in a sound source based on sound source information. Also, it performs a function of temporarily or permanently storing data processed by the processor 130 . Here, the memory 120 may include magnetic storage media or flash storage media, but the scope of the present invention is not limited thereto.

The processor 130, as a kind of central processing unit, controls the entire process of detecting an anomaly in a sound source based on sound source information. Each step performed by the processor 130 will be described later with reference to FIGS. 3 and 4 .

Here, the processor 130 may include all types of devices capable of processing data, such as a processor. Here, a 'processor' may refer to a data processing device embedded in hardware having a physically structured circuit to perform functions expressed by codes or instructions included in a program, for example. As an example of such a data processing device built into hardware, a microprocessor, a central processing unit (CPU), a processor core, a multiprocessor, an application-specific integrated (ASIC) circuit), field programmable gate array (FPGA), etc., but the scope of the present invention is not limited thereto.

The database 140 matches and stores sound sources received from the sound source server 200 and sound source information, and stores classified sound source groups and a plurality of anomaly detection models according to the sound source information.

Although not shown in FIG. 2 , some of data on sound sources, sound source information, sound source groups, and anomaly detection models may be stored in a database (not shown) physically or conceptually separated from the database 140 .

The classification module 150 means a module for classifying and clustering sound sources based on patterns of sound source information. In this case, the classification module 150 may apply at least one of a pattern-specific time-series clustering technique and a clustering technique to cluster the sound source information. However, if a clustering technique with high accuracy is developed according to the development of technology, it can be applied in place of the above two techniques.

The anomaly determination module 160 is a module for determining various states of the sound source based on the generated anomaly detection model. For example, when the classification module 150 generates a sound source group for abnormal streaming or sales status, the abnormality determination module 160 uses an anomaly detection model generated for each type, so that the sound source is similar to a certain sound source group. to judge whether or not

Meanwhile, as an additional embodiment, an anomaly detection model for an abnormal state may not be established because the classification module 150 has not yet generated a sound source group for an abnormal state. At this time, if a specific sound source is not classified into any sound source group (ie, the specific sound source group is not included in the previously learned sound source group, the abnormality determination module 160 needs to generate a new sound source group for classifying the specific sound source group). In this case), it can be determined that a specific sound source is in an abnormal state or a special state. At this time, rather than judging the corresponding sound source as an unconditional abnormal state, the state of a specific sound source is determined in consideration of actual external factors (a state in which the sound source has actually become an issue).

3 is an operation flowchart illustrating a method for detecting an anomaly in a sound source based on sound source information according to an embodiment of the present invention.

Referring to FIG. 3 , the server 100 receives sound source information respectively corresponding to a plurality of sound sources from the sound source server 200 (S110).

At this time, as described above, the sound source information is composed of metadata and log data of the sound source. Specifically, the meta data includes at least one of the title, singer, genre, composer, lyricist, producing, release date, and featuring information of the sound source, and the log data includes information on a sales method of the sound source, information on a payment method, and whether the sound source is streamed or sold. At least one of the number of users who streamed or purchased the sound source is included.

Next, the server 100 creates a sound source group for sound sources (S120).

Specifically, the server 100 generates a plurality of sound source groups by inputting sound source information of a plurality of sound sources into a preset classification model and classifying the type of each sound source.

At this time, as an optional embodiment, the server 100 uses a waveform obtained by calculating meta data and log data according to a predetermined time flow as sound source information.

Accordingly, in step S120, the server 100 generates at least one sound source group by applying any one of pattern-specific time-series clustering analysis or clustering technique to changes in sound source information. For example, referring to FIG. 5 , step-by-step clustering according to changes in sound source information is performed for 12 types of sound sources. At this time, the number of sound source groups to be clustered varies according to the number of clustering criteria (step 1, step 2, step 3, etc.).

At this time, in step S120, the server 100 performs labeling on the type of each classified sound source group based on information (type) having the highest similarity among sound source information included in the sound source group.

Finally, the server 100 detects an abnormal pattern for a specific sound source based on a plurality of anomaly detection models (S130).

Specifically, the server 100 inputs a specific sound source to a preset anomaly detection model corresponding to the type of the sound source group among a plurality of previously generated anomaly detection models by learning a plurality of sound source groups, and determines whether there is an abnormal pattern in the specific sound source. will detect

At this time, as an optional embodiment, the server 100 inputs a group of sound sources with a designated type as an input value, and when a specific sound source is input to the learned anomaly detection model, the type of the specific sound source is included together based on the type learned by the anomaly detection model. can be classified.

As another optional embodiment, in order to determine the abnormal state of the sound source through the sound source classification model generated after step S130, the steps described in FIG. 4 are followed.

4 is an operation flowchart for determining a state of a sound source through an anomaly detection model according to an embodiment of the present invention.

Referring to FIG. 4 , the server 100 inputs specific sound source information to an anomaly detection model (S140).

Next, when a specific sound source is included in the group of abnormal sound sources, the server 100 determines that the specific sound source is in an abnormal state (S150).

Specifically, in step S150, the server 100 determines that the specific sound source is in an abnormal streaming or abnormal sales state when the sound source information of the specific sound source is input to the anomaly detection model and the specific sound source is output as an abnormal sound source. For example, an anomaly detection model that detects abnormal streaming detects that a waveform of sound source information of a specific sound source has an abnormal shape, and identifies that abnormal streaming is in progress.

At this time, the anomaly detection model compares the pattern for the streaming or sale time or number of sound sources corresponding to the corresponding type, and outputs the corresponding sound source as an abnormal sound source group when a different pattern is derived.

On the other hand, as an optional embodiment, in step S150, the server 100 receives the number of mentions of a specific sound source or an issue index indicating the extent to which the corresponding sound source is an issue in the community from an external server including preset SNS and web pages. will do Through this, even if the anomaly detection model of the server 100 outputs a group of abnormal sound sources, if the number of mentions is greater than a preset threshold value, it is determined that the specific sound source is in one of normal streaming or normal sales. For example, a singer who normally has a small amount of sales may attract high attention from the starting point of appearing in a specific broadcast, and the sales volume may temporarily increase than usual. Since this state is not an abnormal state, the server 100 checks what happened to the corresponding singer or sound source and corrects the resulting value.

At this time, the number of mentions or issue index may be crawled and collected from an external server (eg, SNS, comprehensive portal site, search site, etc.). If the status of a specific sound source is changed to a normal streaming or normal sales status, the server 100 creates a new sound source group based on the number of mentions on SNS or the issue index and updates the anomaly detection model.

The above description of the present invention is for illustrative purposes, and those skilled in the art can understand that it can be easily modified into other specific forms without changing the technical spirit or essential features of the present invention. will be. Therefore, the embodiments described above should be understood as illustrative in all respects and not limiting. For example, each component described as a single type may be implemented in a distributed manner, and similarly, components described as distributed may be implemented in a combined form.

The scope of the present invention is indicated by the following claims rather than the detailed description above, and all changes or modifications derived from the meaning and scope of the claims and equivalent concepts should be construed as being included in the scope of the present invention. do.

100: server
200: sound source server

Claims (12)

In the anomaly detection method for a sound source based on sound source information, performed by a server,
(a) receiving sound source information respectively corresponding to a plurality of sound sources from a sound source server;
(b) generating a plurality of sound source groups by inputting the sound source information of the plurality of sound sources into a preset classification model and classifying the type of each sound source; and
(c) A specific sound source is input to a preset anomaly detection model corresponding to the type of the sound source group among a plurality of anomaly detection models pre-generated by learning a plurality of the sound source groups, respectively, to determine whether or not there is an abnormal pattern in the specific sound source. detecting; including,
The sound source information is composed of meta data and log data of the sound source, an anomaly detection method for a sound source based on sound source information. According to claim 1,
The metadata includes at least one of the title, singer, genre, composer, lyricist, producing, release date, and featuring information of the sound source, and the log data includes sales method information, payment method information, and the sound source is streaming. Including at least one of the number of users who streamed or purchased the sound source, the time it was sold or sold,
An anomaly detection method for a sound source based on sound source information, wherein a waveform calculated from the meta data and log data according to a predetermined time flow is used as sound source information. According to claim 2,
The step (b) is
The method of detecting an anomaly for a sound source based on sound source information, wherein at least one sound source group is generated by applying any one of pattern-specific time series clustering or clustering techniques to changes in the sound source information. According to claim 3,
The method of detecting an anomaly with respect to a sound source based on sound source information, further comprising the step of labeling the type of each classified sound source group based on information having the highest similarity among sound source information included in the sound source group. According to claim 4,
The step (c) is
When the specific sound source is input to the anomaly detection model that has been learned by inputting the sound source group with the specified type as an input value, the type of the specific sound source is classified together based on the type learned by the anomaly detection model. An anomaly detection method for a sound source based on information. According to claim 1,
After step (c) above
(d) if the sound source information of the specific sound source is input to the anomaly detection model and the specific sound source is output as an abnormal sound source, determining that the specific sound source is in an abnormal streaming or abnormal sales state; An anomaly detection method for a sound source based on sound source information. According to claim 6,
The anomaly detection model compares the pattern for the streaming or sales time or number of the sound sources corresponding to the corresponding type, and outputs the corresponding sound source as an abnormal sound source group when a different pattern is derived, based on the sound source information. An anomaly detection method for . According to claim 6,
The step (d) is
Receiving the number of mentions of the specific sound source from a preset external server including a preset SNS and web page;
Even if the anomaly detection model is output as an abnormal sound source group, if the number of mentions is greater than a predetermined threshold value, the specific sound source is judged to be in either normal streaming or normal sale state, based on sound source information. anomaly detection method for According to claim 8,
The method of detecting an anomaly for a sound source based on sound source information, wherein the number of mentions is crawled and collected from the external server. According to claim 8,
After step (d) above
When the status of the specific sound source changes to the normal streaming or normal sales status, the sound source group is newly created based on the number of mentions of the SNS and the anomaly detection model is updated. anomaly detection method for An anomaly detection device for a sound source based on sound source information,
A memory storing an anomaly detection program for a sound source based on the sound source information; and
An anomaly detection processor for a sound source based on the sound source information by executing a program stored in the memory;
The processor receives sound source information corresponding to a plurality of sound sources from a sound source server, inputs the sound source information of the plurality of sound sources to a preset classification model, classifies the type of each sound source, and generates a plurality of sound source groups; Detecting whether or not there is an abnormal pattern in the specific sound source by inputting a specific sound source to a preset anomaly detection model corresponding to the type of the sound source group among a plurality of anomaly detection models pre-generated by learning each of the plurality of sound source groups; An anomaly detection device for a sound source based on sound source information, wherein the sound source information is composed of metadata and log data of the sound source. A computer readable storage medium having a program recorded thereon for performing an anomaly detection method for a sound source based on the sound source information according to claim 1.

Images