Method and device for providing private chat service

Abstract

An embodiment of the present invention relates to a method and device for providing a private chat service, wherein message filtering is performed for a large number of users by using a combination of databases allocated to the respective users and a database for storing artist messages, thus making it possible to increase the transmission and reception speed of the private messages.

Description

Method and device for providing private chat service

[0001] The present invention relates to a method and device for providing a private chat service, and more particularly, to a method and device for providing a private chat service for filtering messages between an artist and a plurality of fans following the artist, and increasing the speed of message transmission and reception.

[0002] The recent growth of the entertainment industry and advancements in IT technology have led to a rise in social networking services and fan platforms for communication between celebrities, artists, and fans. These social networking services can serve as fan communities. For example, they can offer comprehensive features such as communication with celebrities and artists, live broadcasts, and fan community functions.

[0003] Furthermore, these fan platforms are expanding their functionality beyond traditional communities, encompassing a wide range of activities, including selling and purchasing merchandise, ticketing concerts and events, providing original content, and implementing non-fungible tokens (NFTs).

[0004] As fan platforms expand their functionality, the notifications and messages provided to users can also vary. For example, if artists can create and post articles, messages, videos, and more directly to the community, users can receive immediate push notifications when artists post or comment, or receive notifications about various benefits for membership subscribers. Furthermore, notifications and messages with different content can be provided depending on the functionality.

[0005] In this way, not only are the functions of fan platforms expanding, but as the entertainment market globalizes, the number of artists, communities, and users on fan platforms is also rapidly increasing. Furthermore, as the influence of the Korean Wave expands user base to a global scale, the difficulty of instantly processing messages between artists and their numerous fans is increasing.

[0006] A method and device for providing a private chat service according to an embodiment of the present invention is intended to increase the speed of message filtering processing in a chat between an artist and a number of fans.

[0007] In addition, the method and device for providing a private chat service according to an embodiment of the present invention is intended to increase the speed of message transmission and reception between an artist and a number of fans.

[0008] In addition, the method and device for providing a private chat service according to an embodiment of the present invention are for providing private messages for each user.

[0009] However, the technical task that this embodiment seeks to achieve is not limited to the technical task described above, and other technical tasks may exist.

[0010] As a technical means for achieving the above-described technical task, a method for providing a private chat service according to an embodiment of the present invention includes the steps of: receiving an artist message from an artist terminal; filtering whether the artist message includes a preset artist-prohibited word; storing the artist message and artist information in a DM-specific database when the artist message does not include the artist-prohibited word; receiving a request for a message update of a private chat from a first fan terminal among the plurality of fan terminals requesting a search for the artist message; generating a private message corresponding to the first fan terminal based on fan account information corresponding to the first fan terminal and the artist message; and transmitting the private message to the first fan terminal, wherein the DM-specific database is configured to store messages on an artist-by-artist basis, and the private message refers to a message in which a specific word or keyword of the message is converted according to a recipient.

[0011] In addition, a private chat method according to an embodiment of the present invention includes the steps of matching a plurality of individual databases corresponding to the plurality of fan terminals, respectively, a step of receiving a fan message from a first fan terminal among the plurality of fan terminals, a step of filtering whether the fan message includes a preset fan taboo word, a step of storing the fan message in a first individual database corresponding to the first fan terminal among the plurality of individual databases, a step of classifying the context of the fan message using a learning-based artificial intelligence model, a step of classifying whether the fan message corresponds to a malicious message based on the classified context, a step of receiving an artist message from the artist terminal, a step of storing the artist message and artist information in a DM-specific database, a step of receiving a message update request of a private chat from the first fan terminal, a step of generating a private message corresponding to the first fan terminal based on fan account information corresponding to the first fan terminal and the artist message, and a step of transmitting the fan message and the private message stored in the first database to the first fan terminal, wherein the DM-specific database is configured to store messages on an artist basis, and the private message refers to a message in which a specific word or keyword of the message is converted according to a recipient.

[0012] In addition, a device for providing a private chat service according to an embodiment of the present invention includes a communication module for transmitting and receiving information between a terminal and the private chat service relay server, a memory for storing a private chat program, a DM-specific database for storing messages by artist, a plurality of individual databases allocated to each user and storing messages by user, and a processor for executing a private chat program stored in the memory, wherein the processor receives a fan message from a first fan terminal among a plurality of fan terminals, stores the fan message in a first individual database corresponding to the first fan terminal among the plurality of individual databases, receives an artist message from the artist terminal, filters whether the fan message and the artist message include preset fan taboo words and artist taboo words, and, if the fan message does not include the fan taboo words, stores the artist message and artist information in the DM-specific database, and, if a request for a message update of a private chat is received from the first fan terminal, generates a private message corresponding to the first fan terminal based on fan account information corresponding to the first fan terminal and the artist message, and transmits the fan message and the private message stored in the first database to the first fan terminal, wherein the private message is It refers to a message in which specific words or keywords in the message are converted according to the recipient.

[0013] A method and device for providing a private chat service according to an embodiment of the present invention can increase the message filtering processing speed in a chat between an artist and a number of fans.

[0014] In addition, the method and device for providing a private chat service according to an embodiment of the present invention can increase the speed of message transmission and reception between an artist and a number of fans.

[0015] In addition, the method and device for providing a private chat service according to an embodiment of the present invention can provide private messages for each user.

[0016] Figure 1 is an exemplary diagram showing a communication connection of a private chat service providing device according to an embodiment of the present invention.

[0017] Figure 2 is a configuration diagram of a private chat service relay server according to an embodiment of the present invention.

[0018] Figure 3 is a configuration diagram of a terminal according to an embodiment of the present invention.

[0019] Figure 4 is a conceptual diagram illustrating the function of a processor according to an embodiment of the present invention.

[0020] FIG. 5 is a data processing concept diagram for fan message storage according to an embodiment of the present invention.

[0021] Figure 6 is a data flow diagram for storing artist messages according to an embodiment of the present invention.

[0022] Figure 7 is a data flow diagram for private chat inquiry of a fan terminal according to an embodiment of the present invention.

[0023] Figure 8 is a data flow diagram for private chat inquiry of a fan terminal according to an embodiment of the present invention.

[0024] Figure 9 is a data flow diagram for private chat inquiry of an artist terminal according to an embodiment of the present invention.

[0025] Figure 10 is a flowchart of a fan message storage method according to an embodiment of the present invention.

[0026] Figure 11 is a flowchart of a method for storing artist messages according to an embodiment of the present invention.

[0027] Figure 12 is a flowchart of a private chat update method according to an embodiment of the present invention.

[0028] Figure 13 is a flowchart of a private chat update method according to an embodiment of the present invention.

[0029] Figure 14 is a flowchart of a private chat update method according to an embodiment of the present invention.

[0030] Figure 15 is a conceptual diagram illustrating the function of a processor according to an embodiment of the present invention.

[0031] Figure 16 is a data flow diagram for fan message filtering according to an embodiment of the present invention.

[0032] Figure 17 is a data flow diagram for artist message filtering according to an embodiment of the present invention.

[0033] Figure 18 is a flowchart of a fan message filtering method according to an embodiment of the present invention.

[0034] Figure 19 is a flowchart of an artist message filtering method according to an embodiment of the present invention.

[0035] Below, with reference to the attached drawings, embodiments of the present invention are described in detail so that those skilled in the art can easily implement them. However, the present invention may be implemented in various different forms and is not limited to the embodiments described herein. In the drawings, irrelevant parts have been omitted for clarity of description, and similar reference numerals have been used throughout the specification to indicate similar elements.

[0036] Throughout the specification, when a part is said to be "connected" to another part, this includes not only the cases where the parts are "directly connected" but also the cases where the parts are "electrically connected" with other elements intervening. Furthermore, when a part is said to "include" a component, this does not exclude other components, but rather includes other components, unless otherwise stated.

[0037] In addition, the attached drawings are only intended to facilitate easy understanding of the embodiments disclosed in this specification, and the technical ideas disclosed in this specification are not limited by the attached drawings, and should be understood to include all modifications, equivalents, or substitutes included in the spirit and technical scope of the present invention.

[0038] Terms that include ordinal numbers, such as first, second, etc., may be used to describe various components, but the components are not limited by these terms. These terms are used solely to distinguish one component from another.

[0039] When a component is referred to as being "connected" or "connected" to another component, it should be understood that it may be directly connected or connected to that other component, but that there may be other components intervening. Conversely, when a component is referred to as being "directly connected" or "connected" to another component, it should be understood that there are no other components intervening.

[0040] Singular expressions include plural expressions unless the context clearly indicates otherwise.

[0041] In this application, terms such as “include” or “have” are intended to specify the presence of a feature, number, step, operation, component, part or combination thereof described in the specification, but should be understood not to exclude in advance the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts or combinations thereof.

[0042] Hereinafter, a device for providing a private chat service according to an embodiment of the present invention will be described with reference to FIG. 1.

[0043] FIG. 1 is an exemplary diagram showing a communication connection between a private chat service relay server (100) and a terminal (200) according to an embodiment of the present invention.

[0044] Referring to FIG. 1, a private chat service relay server (100) is connected to a terminal (200) via a communication network. At this time, the private chat service relay server (100) refers to a device for providing message transmission and reception between users using a fan platform, social network service, etc., and may correspond to a fan platform server, a social network service server, etc. The terminal (200) includes an artist terminal (200-1) used by celebrities, artists, etc., and a fan terminal (200-2) used by fans following celebrities, artists, etc. The terminal (200) may refer to a device that displays messages and notifications received from the private chat service relay server (100) to users.

[0045] The private chat service relay server (100) can generate and provide users with a list of messages and notifications on fan platforms, social network services, and the like. Rather than simply providing users with messages and notifications, the server can provide users with personalized notification messages and notification lists based on user-specific filtering information. Furthermore, the private chat service relay server (100) can build individual databases for each user and store messages individually. The specific configuration of the private chat service relay server (100) will be described in detail with reference to FIGS. 2 and 4 , which will be described later.

[0046] The terminal (200) may refer to any type of handheld-based wireless communication device such as a notebook, desktop, laptop, wireless communication device with portability and mobility, or a smartphone, tablet PC, etc. equipped with a web browser.

[0047] In addition, the communication network illustrated in FIG. 1 can be implemented as a wired network such as a Local Area Network (LAN), a Wide Area Network (WAN), or a Value Added Network (VAN), or any type of wireless network such as a mobile radio communication network or a satellite communication network.

[0048] Hereinafter, the structure of a private chat service relay server according to an embodiment of the present invention will be described with reference to FIG. 2.

[0049] FIG. 2 is a structural diagram illustrating the structure of a private chat service relay server (100) according to an embodiment of the present invention.

[0050] Referring to FIG. 2, the private chat service relay server (100) includes a communication module (110), a memory (120), and a processor (140), and may further include a database (130). The communication module (110) performs information transmission and reception with a terminal (200). The communication module (110) may include a device including hardware and software necessary for transmitting and receiving signals, such as control signals or data signals, using wired or wireless connections with other network devices.

[0051] The memory (120) stores a private chat program. The name of the private chat program is set for convenience of explanation, and the name itself does not limit the program's functions. The memory (120) can store at least one of the following: information and data input to the communication module (110), information and data required for functions performed by the processor (140), and data generated by the execution of the processor (140).

[0052] Memory (120) should be interpreted as a general term for a non-volatile storage device that maintains stored information even when no power is supplied, and a volatile storage device that requires power to maintain the stored information. In addition, memory (120) may perform a function of temporarily or permanently storing data processed by the processor (140). Memory (120) may include magnetic storage media or flash storage media in addition to volatile storage devices that require power to maintain stored information, but the scope of the present invention is not limited thereto.

[0053] The database (130) can store data related to artist information, user information, subscription information, artist messages, fan messages, and private messages that are converted and provided for each fan user. The database (130) may constitute a portion of the memory (120), but is not necessarily located within the private chat service relay server (100). Rather, it may be connected to the outside of the private chat service relay server (100) to transmit and receive data using a communication connection.

[0054] Additionally, the database (130) can be configured to store message data on an artist-by-artist and user-by-user basis. The detailed configuration of the database (130) will be described in detail with reference to FIG. 4 described below.

[0055] The processor (140) is configured to execute a private chat program in the memory (120). The processor (140) may include various types of devices that control and process data. The processor (140) may refer to a data processing device built into hardware that has a physically structured circuit to perform a function expressed by a code or command included in the program. In one example, the processor (140) may be implemented in the form of a microprocessor, a central processing unit (CPU), a processor core, a multiprocessor, an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), etc., but the scope of the present invention is not limited thereto.

[0056] The processor (140) is configured to execute a private chat program and perform the following functions and procedures.

[0057] Referring to FIGS. 1 and 2 together, the processor (140) receives an artist message generated from an artist terminal (200-1) and transmits the artist message to a fan terminal (200-2) that follows the artist. In addition, the processor (140) may receive a fan message generated from a fan terminal (200-2) and provide the fan message to the artist terminal (200-1).

[0058] Additionally, the processor (140) may generate private messages, in which the content of the artist message is converted for each fan user, and provide the converted private messages to the fan terminal (200-2). At this time, to increase the message conversion speed for multiple fan users, the processor (140) may generate private messages for each fan user and store the private messages in a user-specific database. Accordingly, even if multiple users utilize the private chat service, private messages can be provided immediately. Furthermore, the private chat service can be provided at an increased speed compared to generating and providing private messages in response to a user's message inquiry request.

[0059] Additionally, the processor (140) can store artist messages in the database (130), and while private messages are being created and stored in the user-specific database, the processor (140) can read the artist messages from the database (130) to create and provide private messages. After the private messages are stored in the user-specific database, the processor (140) can delete duplicate artist messages and receive message information stored in the user-specific database to provide a private chat service to the user.

[0060] Figure 3 is a block diagram illustrating the configuration of a terminal (200).

[0061] Referring to FIG. 3, the terminal (200) includes a memory (220), an input / output module (230), and a processor (240), and may further include a communication module (210).

[0062] The communication module (210) can transmit and receive information with an external database or external device. Here, the external device may be the private chat service relay server (100 in FIG. 1) described above. The memory (220) stores a private chat program. The name of the private chat program is set for convenience of explanation, and the name itself does not limit the function of the program. The input / output module (230) can receive information, data, etc. transmitted from the outside to the terminal (200), or output information, data, etc. held by the terminal (200) to the outside. For example, the input / output module (230) may include a display, a touchpad, etc. The processor (240) executes the private chat program stored in the memory (220). Additional descriptions of the communication module (210), memory (220), and processor (240) are replaced with the descriptions of the communication module (110 of FIG. 2), memory (120 of FIG. 2), and processor (140 of FIG. 2) previously described with reference to FIG. 2.

[0063] The processor (240) is configured to execute a private chat program to perform the following functions and procedures.

[0064] The processor (240) can transmit artist messages or fan messages generated in the terminal (200) to the private chat service relay server (100). In addition, the processor (240) can receive artist messages, fan messages, private messages, feeds, alarms, etc. stored and generated in the private chat service relay server (100).

[0065] Hereinafter, the operation of the processor (140) and database (130) according to an embodiment of the present invention will be described in detail with reference to FIG. 4.

[0066] FIG. 4 is a conceptual diagram illustrating the function of a private chat service relay server according to an embodiment of the present invention.

[0067] Referring to FIG. 4, the processor (140) may be configured to perform the functions of a control unit (141) and a message queue unit (not shown) described below. In addition, the database (130) may be composed of a DM (Direct Message) database (1310), an individual database (1320), and a sub-database (1330).

[0068] The control unit (141) stores artist messages and fan messages generated from the artist terminal (200-1) and fan terminal (200-2) in the database (130), or performs the function of a controller for reading messages from the database (130) for private chat updates. The message queue unit (not shown) performs a message issuing function for issuing artist messages or fan messages stored in the database (130) to recipients.

[0069] The DM database (1310) can store messages by private chat unit or artist unit. In this case, the private chat unit may refer to a chat room unit where messages are exchanged between fan users and artists. The individual database (1320) can be constructed for each user, allowing messages to be stored by user. Accordingly, individual databases are created, from the first individual database (1321) to the Nth individual database (1321), to correspond to the number of users.

[0070] Additionally, the individual database (1320) may be formed to include an artist transmission database (1322) for storing messages sent by the artist and an artist reception database (1323) for storing messages received by the artist.

[0071] The sub-database (133) may be formed to include a subscription information storage unit (1331) that stores information about fan users following the artist, and a message body storage unit (1332) that stores the contents of artist messages and fan messages.

[0072] Below, the interaction between the processor (140) and the database (130) that occurs when providing a private chat service between an artist and a fan is described in detail.

[0073] The processor (140) can receive a fan message generated from a first fan terminal among a plurality of fan terminals (200-2). The processor (140) controls the fan message to be stored in a first individual database corresponding to the first fan terminal among a plurality of individual databases (1320). In addition, when a fan message is stored in the first individual database, the processor (140) can generate a message feed for the fan message and provide it to the artist terminal (200-1).

[0074] Additionally, the processor (140) can receive an artist message from the artist terminal (200-1). When the processor (140) receives an artist message, it can control the storage of the artist message and artist information in the DM database (1310). When the artist message is stored in the DM database (1310), the processor (140) generates a message feed provided to multiple fan terminals and transmits the message feed to the first fan terminal.

[0075] When a first fan terminal requests a private chat update, the processor (140) receives a private chat message update request from the first fan terminal. Then, the processor (140) generates a private message corresponding to the first fan terminal based on fan account information and artist messages corresponding to the first fan terminal. The processor (140) transmits the fan message stored in the first database and the generated private message to the first fan terminal. At this time, the private message may refer to a message in which specific words or keywords in the message are converted according to the recipient. Furthermore, the message feed may include notifications, posts, etc. to notify the recipient of the message of the receipt of the message.

[0076] Additionally, when the processor (140) receives an artist message, it can store the artist message in a DM-specific database (1310) and at the same time convert the artist message into a private message and store it in an individual database (1320).

[0077] Specifically, the processor (140) generates a plurality of private messages corresponding to each of the plurality of fan terminals based on fan account information and artist messages corresponding to each of the plurality of fan terminals. Then, the processor (140) stores the plurality of private messages in a corresponding individual database based on the fan account information. When the private messages are completely stored in the individual database, the processor (140) controls to delete the artist messages stored in the DM-specific database (1310). In addition, after the private messages are stored in the individual database, the processor receives the private messages from the individual database (1320) rather than the DM-specific database (1310) and transmits them to the fan terminal.

[0078] That is, when an artist writes a message, the private chat service relay server (100) generates a corresponding private message for each subscriber fan and stores the private message in an individual database assigned to each user. Therefore, when a user requests a message inquiry or update, the message is retrieved from the individual database, enabling the private chat service to be quickly provided to multiple users.

[0079] In addition, the processor (140) stores artist messages and artist information in a DM-specific database (1310), and when a preset standard time elapses, deletes the artist messages stored in the DM-specific database (1310), receives a private message from an individual data bay (1320), and transmits the private message to a fan terminal.

[0080] At this time, the DM database (1310) may be formed based on NoSQL (Not Only SQL), and the individual database (1320) may be formed based on SQL (Structured query language).

[0081] Accordingly, the DM-specific database (1310) has a non-relational database structure and can store unstructured data. In addition, the individual database (1320) has a relational database structure and can create a DB (DataBase) table for each user and store message data for each user by pre-applying a user-specific DB table filter.

[0082] Since the individual database (1320) is configured using SQL, its write speed is slower and its read speed is faster than that of the DM-specific database (1310). On the other hand, the DM-specific database (1310) has a faster write speed and a slower read speed than that of the individual database (1320).

[0083] The private chat service relay server (100) can increase the speed of providing private chat services to multiple users by providing private messages stored using individual databases (1320). Furthermore, while private messages are stored in individual databases (1320), the DM-specific database (1310) can be used to provide private messages, thereby supplementing the operation during the time messages are stored in individual databases.

[0084] Hereinafter, the specific data processing process between the processor (140) and the database (130) will be described in detail with reference to FIGS. 5 to 9 described below.

[0085] FIG. 5 is a data flow diagram for storing fan messages according to an embodiment of the present invention.

[0086] Referring to FIG. 5, when a fan user writes and sends a fan message to be sent to an artist, fan message data is transmitted from a fan terminal (200-2) to the control unit (141) of the private chat service relay server (100). The fan message data may include at least one of the following: message content, information on the artist receiving the message, and the fan's subscription information on the artist.

[0087] The control unit (141) transmits the received fan message data to the message body storage unit (1332) and requests that the fan message data be stored in the message body storage unit (1332). The message body storage unit (1332) stores the fan message data, and when the storage of the fan message data is completed, generates a fan message ID corresponding to the fan message and responds the fan message ID to the control unit (141).

[0088] The control unit (141) that receives the fan message ID requests the storage of fan message data in the individual database (1320). At this time, the individual database (1320) refers to a database assigned to each user. Accordingly, when a fan message is received in the first fan terminal, the storage of fan message data is requested in the first individual database (1321) assigned to the first fan terminal. In addition, the fan message data stored in the individual database (1320) may include the fan message ID and subscription information.

[0089] When fan message data is stored in the individual database (1320), the individual database (1320) transmits a fan message data storage completion response to the control unit (141).

[0090] The control unit (141), which has received a response from the individual database (1320) regarding the completion of fan message data storage, requests the subscription information storage unit (1331) to retrieve the fan's subscription information. Specifically, a message queue unit (not shown) included in the control unit (141) requests the subscription information storage unit (1331) to retrieve the fan's subscription information based on the fan's subscriber ID information.

[0091] The subscription information storage unit (1331) that receives a subscription information inquiry request responds to the control unit (141) with a DM (Direct Message) ID corresponding to the subscriber ID information. At this time, the DM ID may correspond to a private chat room ID or artist information.

[0092] The control unit (141) that receives the DM ID from the subscription information storage unit (1331) requests the storage of fan message data in the artist reception database (1323) of the target artist of the fan message based on the DM ID. At this time, the fan message data stored in the artist reception database (1323) may include fan message ID and DM ID information.

[0093] When the storage of fan message data in the artist reception database (1323) is completed, the artist reception database (1323) responds to a fan message issuance request to the control unit (141).

[0094] The control unit (141) that receives a request to issue a fan message from the artist reception database (1323) creates a fan feed for the artist to be sent to the target artist of the fan message, and transmits the fan feed for the artist to the terminal (200) of the target artist.

[0095] Accordingly, when a fan sends a message to an artist, the fan's message is stored in the artist's reception database. Furthermore, when the artist wishes to check the artist's fan feed to view fan messages, the private chat relay server (100) can read the fan message based on the fan message data stored in the artist reception database and provide it to the artist's terminal.

[0096] Below, the data processing sequence for storing artist messages is described with reference to FIG. 6.

[0097] Figure 6 is a data flow diagram for storing artist messages according to an embodiment of the present invention.

[0098] Referring to FIG. 6, when an artist user writes and transmits an artist message to be transmitted to a fan user, artist message data is transmitted from the artist terminal (200-1) to the control unit (141) of the private chat service relay server (100). The artist message data may include at least one of message content, message recipient information, and DM ID.

[0099] The control unit (141) transmits the received artist message data to the message body storage unit (1332) and requests that the fan message data be stored in the message body storage unit (1332). The message body storage unit (1332) stores the artist message data, and when the storage of the artist message data is completed, generates an artist message ID corresponding to the artist message and responds the artist message ID to the control unit (141).

[0100] The control unit (141) that receives the artist message ID requests the artist message data to be stored in the artist transmission database (1322). At this time, the artist transmission database (1322) is an individual database assigned to the artist, and refers to a database in which messages written by the artist to be transmitted are stored. In addition, the artist transmission database (1322) may be configured in combination with the artist reception database (1323) or may be configured separately. In addition, the artist message data stored in the artist reception database (1322) may include data regarding at least one of the artist message ID and the DM ID.

[0101] When artist message data is stored in the artist reception database (1322), the artist reception database (1322) transmits a response indicating completion of artist message data storage to the control unit (141).

[0102] The control unit (141), which has received a response indicating completion of artist message data storage from the artist reception database (1322), requests the storage of artist message data in the DM-specific database (1310). At this time, the artist message data stored in the DM-specific database (1320) includes data regarding at least one of the artist message ID and the DM ID.

[0103] When the storage of artist message data in the DM database (1310) is completed, the DM database (1310) transmits a response indicating completion of artist message data storage to the control unit (141).

[0104] The control unit (141), which has received a response from the DM database (1310) indicating completion of storing artist message data, requests the subscription information storage unit (1331) to retrieve subscriber list information corresponding to the DM ID. Specifically, a message queue unit (not shown) included in the control unit (141) requests the subscription information storage unit (1331) to retrieve a list of subscriber IDs that subscribe to the artist based on the DM ID.

[0105] When the control unit (141) receives a subscriber ID list from the subscription information storage unit (1331), the control unit generates a private message for each subscriber based on the artist message data and the subscriber ID list. In addition, the control unit (141) requests the storage of the corresponding private message in the individual database (1321) of each subscriber.

[0106] At this time, requesting storage of a private message in an individual database (1321) may mean that the control unit (141) stores the contents of the private message in the message body storage unit (1332) and requests storage of data for one or more of the private message ID, DM ID, and subscriber ID received in response from the message body storage unit (1332).

[0107] Accordingly, when an artist sends a message to a fan, a private message for each fan is stored in the fan's individual database (1321). Additionally, when a fan wishes to view the message, the private chat relay server (100) can read the private message data stored in the individual database (1321) and provide it to the fan terminal.

[0108] Below, with reference to FIG. 7, the data processing process when a fan user requests to view a private chat before the private message is stored in an individual database is described in detail.

[0109] Figure 7 is a data flow diagram when a fan searches for a message according to an embodiment of the present invention.

[0110] Referring to FIG. 7, when a fan requests to view a private chat or artist message, the control unit (141) receives a private chat view request from the fan terminal (200-2). At this time, the private chat view request includes the fan's subscriber ID information.

[0111] When the control unit (141) receives a private chat inquiry request from a fan terminal (200-2), the control unit (141) requests a message inquiry corresponding to the subscriber ID to the individual database (1320) corresponding to the fan terminal (200-2) or fan account.

[0112] The individual database (1320) may respond with a list of stored message IDs in response to a message query request. At this time, the message that the individual database (1320) responds to the control unit (141) may include at least one of a fan message ID and a private message ID. However, before the private message is stored in the individual database (1320), the message ID that the individual database (1320) responds to the control unit (141) includes only a fan message ID.

[0113] Additionally, the control unit (141) requests the subscription information storage unit (1331) to search for a DM ID corresponding to the fan's subscriber ID. The subscription information storage unit (1331) that receives the DM ID search request responds to the control unit (141) with the DM ID stored corresponding to the subscriber ID.

[0114] The control unit (141) that receives the DM ID requests the DM database (1310) to search for a message corresponding to the DM ID. The DM database (1310) that receives the message search request responds to the control unit (141) with the artist message ID stored corresponding to the DM ID.

[0115] Accordingly, the control unit (141) receives a fan message ID from an individual database (1320) and an artist message ID from a DM-specific database (1310). Then, the control unit (141) requests message body information corresponding to the fan message ID and artist message ID from the message body storage unit (1332) based on the received fan message ID and artist message ID. Accordingly, the message body storage unit (1332) responds with message information corresponding to the received fan message ID and artist message ID.

[0116] The control unit (141) can transmit received fan messages and artist messages to the fan terminal (200-2). Additionally, the control unit (141) can convert artist messages into private messages based on the fan's subscription information and transmit them to the fan terminal (200-2).

[0117] Below, with reference to FIG. 8, the data processing process when a fan user requests to view a private chat after a private message has been stored in an individual database is described in detail.

[0118] Figure 8 is a data flow diagram when a fan searches for a message according to an embodiment of the present invention.

[0119] Referring to FIG. 8, when the private message storage in the individual database (1320) is completed, the control unit (141) requests the deletion of the completed artist message in the DM database (1310). The completed artist message refers to an artist message or artist message ID converted into a private message.

[0120] By deleting the artist message ID stored in the DM star database (1310), it is possible to prevent duplicate artist messages and private messages from being sent to fan terminals.

[0121] After the deletion of the artist message ID from the DM database (1310) is completed, when a request is made to search for a private chat or artist message from a fan terminal (200), the control unit (141) receives a private chat search request from the fan terminal (200-2). At this time, the private chat search request includes the fan's subscriber ID information.

[0122] When the control unit (141) receives a private chat inquiry request from a fan terminal (200-2), the control unit (141) requests an individual database (1320) corresponding to the fan terminal (200-2) or fan account (200-2) to look up a message corresponding to the subscriber ID.

[0123] The individual database (1320) may respond with a list of stored message IDs in response to a message query request. In this case, the message that the individual database (1320) responds to the control unit (141) may include both a fan message ID and a private message ID.

[0124] Accordingly, the control unit (141) receives a fan message ID and a private message ID from the individual database (1320). Then, the control unit (141) requests the message body storage unit (1332) for message body information corresponding to the fan message ID and the private message ID based on the received fan message ID and private message ID. Accordingly, the message body storage unit (1332) responds with message information corresponding to the received fan message ID and private message ID.

[0125] The control unit (141) can transmit received fan messages and private messages to the fan terminal (200-2). Therefore, even if a plurality of fan terminals (200-2) request a private chat inquiry, the private chat service relay server (100) only reads messages from the individual database (1320) corresponding to the fan terminal (200-2). Accordingly, even if a plurality of users request a message inquiry, a fast processing speed can be maintained.

[0126] Below, with reference to Figure 9, the data processing process when an artist requests to view a private chat is described in detail.

[0127] Figure 9 is a data flow diagram when searching for an artist's message according to an embodiment of the present invention.

[0128] Referring to FIG. 9, when an artist requests to view a private chat, the control unit (141) receives a private chat view request from the artist terminal (200-1). At this time, the private chat view request includes DM ID information.

[0129] When the control unit (141) receives a private chat inquiry request from the artist terminal (200-2), the control unit (141) requests the artist transmission database (1322) to look up a message corresponding to the DM ID. Accordingly, the artist transmission database (1322) responds with a list of artist message IDs stored corresponding to the artist's DM ID.

[0130] Additionally, the control unit (141) requests the artist reception database (1323) to search for a message corresponding to the DM ID. Accordingly, the artist reception database (1323) responds with a list of fan message IDs stored corresponding to the artist's DM ID.

[0131] Accordingly, the control unit (141) receives an artist message ID list from the artist transmission database (1322) and a fan message ID list from the artist reception database (1323). The control unit (141) requests a message information inquiry from the message body storage unit (1332) based on the received artist message ID list and fan message ID list.

[0132] Accordingly, the message body storage unit (1332) responds to the control unit (141) with message information corresponding to the artist message ID list and the fan message ID list. The control unit (141) that receives the artist message and fan message information updates the message content to the artist terminal (200).

[0133] Hereinafter, a method for providing a private chat service will be described in detail with reference to FIGS. 10 to 14.

[0134] Figure 10 is a flowchart of a fan message storage method according to an embodiment of the present invention.

[0135] Referring to FIG. 10, a method for storing fan messages sent by a fan user to an artist in a private chat service according to an embodiment of the present invention includes a fan message receiving step (S110), a fan message storing step (S120), and a message feed providing step (S130).

[0136] Specifically, in the fan message reception step (S110), when a fan user writes and transmits a fan message to be sent to an artist, fan message data is transmitted from the fan terminal (200-2) to the control unit (141) of the private chat service relay server (100). The fan message data may include at least one of the message content, information on the artist receiving the message, and the fan's artist subscription information.

[0137] In the fan message storage step (S120), the control unit (141) stores the received fan message data in an individual database (1320). Specifically, the control unit (141) transmits the received fan message data to the message body storage unit (1332) and requests that the fan message data be stored in the message body storage unit (1332). Then, the message body storage unit (1332) stores the fan message data, and when the storage of the fan message data is completed, generates a fan message ID corresponding to the fan message and responds the fan message ID to the control unit (141).

[0138] The control unit (141) that receives the fan message ID requests the storage of fan message data in the individual database (1320). At this time, the individual database (1320) refers to a database assigned to each user. Accordingly, when a fan message is received in the first fan terminal, the storage of fan message data is requested in the first individual database (1321) assigned to the first fan terminal. In addition, the fan message data stored in the individual database (1320) may include the fan message ID and subscription information.

[0139] In the message feed provision step (S130), the control unit (141) generates a message feed to be transmitted to the artist who is the recipient of the fan message and transmits it to the artist terminal (200). Specifically, when fan message data is stored in the individual database (1320), the individual database (1320) transmits a fan message data storage completion response to the control unit (141).

[0140] The control unit (141), which has received a response from the individual database (1320) regarding the completion of fan message data storage, requests the subscription information storage unit (1331) to retrieve the fan's subscription information. Specifically, a message queue unit (not shown) included in the control unit (141) requests the subscription information storage unit (1331) to retrieve the fan's subscription information based on the fan's subscriber ID information.

[0141] The subscription information storage unit (1331) that receives a subscription information inquiry request responds to the control unit (141) with a DM (Direct Message) ID corresponding to the subscriber ID information. At this time, the DM ID may correspond to a private chat room ID or artist information.

[0142] The control unit (141) that receives the DM ID from the subscription information storage unit (1331) requests the storage of fan message data in the artist reception database (1323) of the target artist of the fan message based on the DM ID. At this time, the fan message data stored in the artist reception database (1323) may include fan message ID and DM ID information.

[0143] When the storage of fan message data in the artist reception database (1323) is completed, the artist reception database (1323) responds to a fan message issuance request to the control unit (141).

[0144] The control unit (141) that receives a request to issue a fan message from the artist reception database (1323) creates a fan feed for the artist to be sent to the target artist of the fan message, and transmits the fan feed for the artist to the terminal (200) of the target artist.

[0145] Accordingly, when a fan sends a message to an artist, the fan's message is stored in the artist's reception database. Furthermore, when the artist wishes to check the artist's fan feed to view fan messages, the private chat relay server (100) can read the fan message based on the fan message data stored in the artist reception database and provide it to the artist's terminal.

[0146] Hereinafter, a method for saving artist messages will be described with reference to FIG. 11.

[0147] Figure 11 is a flowchart of a method for storing artist messages according to an embodiment of the present invention.

[0148] Referring to FIG. 11, the method for storing artist messages includes an artist message receiving step (S210), an artist message storing step in an individual database (S220), an artist message storing step in a DM-specific database (S230), a private message generating step (S240), and a private message storing step (S250).

[0149] When an artist user writes and transmits an artist message to be transmitted to a fan user in the artist message reception step (S210), artist message data is transmitted from the artist terminal (200-1) to the control unit (141) of the private chat service relay server (100). The artist message data may include at least one of message content, message recipient information, and DM ID.

[0150] In the step of storing artist messages in individual databases (S220), the control unit (141) stores the received artist message data in the artist transmission database (1322) among the artist's individual databases (1320).

[0151] Specifically, the control unit (141) transmits the received artist message data to the message body storage unit (1332) and requests that the fan message data be stored in the message body storage unit (1332). The message body storage unit (1332) stores the artist message data, and when the storage of the artist message data is completed, generates an artist message ID corresponding to the artist message and responds the artist message ID to the control unit (141).

[0152] The control unit (141) that receives the artist message ID requests the artist message data to be stored in the artist transmission database (1322). At this time, the artist transmission database (1322) is an individual database assigned to the artist, and refers to a database in which messages written by the artist to be transmitted are stored. In addition, the artist transmission database (1322) may be configured in combination with the artist reception database (1323) or may be configured separately. In addition, the artist message data stored in the artist reception database (1322) may include data regarding at least one of the artist message ID and the DM ID.

[0153] When artist message data is stored in the artist reception database (1322), the artist reception database (1322) transmits a response indicating completion of artist message data storage to the control unit (141).

[0154] In the step of saving artist messages in the DM-specific database (S230), the control unit (141) saves the received artist messages in the DM-specific database (1310).

[0155] Specifically, the control unit (141) requests the storage of artist message data in the DM-specific database (1310). The DM-specific database (1310) stores artist messages or message IDs for each DM ID. At this time, the artist message data stored in the DM-specific database (1320) includes data regarding at least one of the artist message ID and the DM ID.

[0156] In the private message generation step (S240), the control unit (141) generates a private message to be transmitted to a fan user who subscribes to the artist based on the received artist message.

[0157] Specifically, the control unit (141) that has received a response indicating completion of artist message data storage from the DM database (1310) requests the subscription information storage unit (1331) to retrieve subscriber list information corresponding to the DM ID. Specifically, a message queue unit (not shown) included in the control unit (141) requests the subscription information storage unit (1331) to retrieve a list of subscriber IDs that subscribe to the artist based on the DM ID.

[0158] When the control unit (141) receives a subscriber ID list from the subscription information storage unit (1331), the control unit generates a private message for each subscriber based on the artist message data and the subscriber ID list.

[0159] In the private message storage step (S250), the control unit (141) stores the generated private message in a corresponding individual database (1320).

[0160] Specifically, private messages are generated for each fan user who subscribes to the artist. Furthermore, each fan user is assigned a separate database (1321). Accordingly, the control unit (141) stores the generated private messages in each fan user's individual database (1321).

[0161] At this time, requesting storage of a private message in an individual database (1321) may mean that the control unit (141) stores the contents of the private message in the message body storage unit (1332) and requests storage of data for one or more of the private message ID, DM ID, and subscriber ID received in response from the message body storage unit (1332).

[0162] Accordingly, when an artist sends a message to a fan, a private message for each fan is stored in the fan's individual database (1321). Additionally, when a fan wishes to view the message, the private chat relay server (100) can read the private message data stored in the individual database (1321) and provide it to the fan terminal.

[0163] Hereinafter, a method for updating a private chat for a fan user will be described with reference to FIGS. 12 and 13.

[0164] Figure 12 is a flowchart of a private chat update method of a fan terminal according to an embodiment of the present invention.

[0165] Referring to FIG. 12, the method for updating a fan's private chat and updating a private chat according to a request for inquiry includes a step of receiving a private chat update request (S310), a step of confirming storage of a private message (S320), a step of searching an individual database (S340), a step of searching a database by DM (S332), and a step of updating a private chat (S350).

[0166] In the private chat update request reception step (S310), when a fan requests to view a private chat or artist message, the control unit (141) receives a private chat inquiry request from the fan terminal (200-2). At this time, the private chat inquiry request includes the fan's subscriber ID information.

[0167] In the private message storage confirmation step (S320), it is confirmed whether the private message is being stored in the individual database (1321) or whether storage has been completed.

[0168] While private messages are stored in individual databases (1321), artist messages are read from the DM database rather than the individual database to update the private chat.

[0169] Specifically, if the private message is not stored in an individual database (1321), an individual database lookup step (S331), a DM-specific database lookup step (S332), and a private message creation step (S333) are performed.

[0170] In the individual database query step (S331), the control unit (141) requests a message query corresponding to the subscriber ID to the individual database (1320) corresponding to the fan terminal (200-2) or fan account (200-2).

[0171] The individual database (1320) may respond with a list of stored message IDs in response to a message query request. At this time, the message that the individual database (1320) responds to the control unit (141) may include at least one of a fan message ID and a private message ID. However, before the private message is stored in the individual database (1320), the message ID that the individual database (1320) responds to the control unit (141) includes only the fan message ID. Therefore, in the individual database query step (S331), fan message data written by the fan can be received.

[0172] In the DM database query step (S332), the control unit (141) requests the subscription information storage unit (1331) to search for a DM ID corresponding to the fan's subscriber ID. The subscription information storage unit (1331) that receives the DM ID query request responds to the control unit (141) with the DM ID stored corresponding to the subscriber ID.

[0173] The control unit (141) that receives the DM ID requests the DM database (1310) to search for a message corresponding to the DM ID. The DM database (1310) that receives the message search request responds to the control unit (141) with the artist message ID stored corresponding to the DM ID.

[0174] Accordingly, the control unit (141) receives a fan message ID from an individual database (1320) and an artist message ID from a DM-specific database (1310). Then, the control unit (141) requests message body information corresponding to the fan message ID and artist message ID from the message body storage unit (1332) based on the received fan message ID and artist message ID. Accordingly, the message body storage unit (1332) responds with message information corresponding to the received fan message ID and artist message ID.

[0175] The control unit (141) can transmit received fan messages and artist messages to the fan terminal (200-2).

[0176] Additionally, in the private message generation step (S333), the control unit (141) can convert the artist message into a private message based on the fan's subscription information and transmit it to the fan terminal (200-2).

[0177] If the private message storage in the individual database (1320) is completed in the private message storage confirmation step (S320), only the individual database query step (S340) is performed to update the fan's private chat.

[0178] In the individual database query step (S340), the control unit (141) requests the deletion of a completed artist message from the DM database (1310). A completed artist message refers to an artist message or artist message ID converted into a private message.

[0179] By deleting the artist message ID stored in the DM star database (1310), it is possible to prevent duplicate artist messages and private messages from being sent to fan terminals.

[0180] After the deletion of the artist message ID from the DM database (1310) is completed, when a request is made to search for a private chat or artist message from a fan terminal (200), in the individual database search step (S340), the control unit (141) requests a search for a message corresponding to the subscriber ID from the individual database (1320) corresponding to the fan terminal (200-2) or fan account.

[0181] The individual database (1320) may respond with a list of stored message IDs in response to a message query request. In this case, the message that the individual database (1320) responds to the control unit (141) may include both a fan message ID and a private message ID.

[0182] Accordingly, the control unit (141) receives a fan message ID and a private message ID from the individual database (1320). Then, the control unit (141) requests the message body storage unit (1332) for message body information corresponding to the fan message ID and the private message ID based on the received fan message ID and private message ID. Accordingly, the message body storage unit (1332) responds with message information corresponding to the received fan message ID and private message ID.

[0183] In the private chat update step (S350), the control unit (141) transmits the received fan message and private message to the fan terminal (200-2) to update the private chat of the fan terminal (200-2).

[0184] Accordingly, even if a private chat query is requested from multiple fan terminals (200-2), the private chat service relay server (100) only reads messages from the individual database (1320) corresponding to each fan terminal (200-2). Accordingly, even if multiple users request message query, a fast processing speed can be maintained.

[0185] Additionally, the method for updating a private chat of a fan terminal according to an embodiment of the present invention can be applied by changing the private chat storage confirmation step (S320) to a reference time elapsed confirmation step (S321).

[0186] In the step of checking the elapsed time (S321), the control unit (141) determines that the private message is stored in the individual database (1321) before the preset standard time elapses. Accordingly, if the fan terminal (200) requests an update of the private chat before the preset standard time elapses after receiving the artist message, the control unit (141) performs the individual database search step (S331), the DM-specific database search step (S332), and the private message creation step (S333).

[0187] That is, the control unit (141) retrieves artist message information stored in the DM database (1310) before a preset standard time elapses after receiving the artist message, and creates a private message based on the artist message and provides it to the fan terminal (200).

[0188] In addition, in the step of checking the elapsed time (S321), the control unit (141) determines that the private message has been completely stored in the individual database (1321) if the preset standard time has elapsed. Accordingly, the control unit (141) receives the artist message and, after the preset time has elapsed, performs the step of receiving the message in the individual database (S340) if the fan terminal (200) requests an update of the private chat.

[0189] That is, after a preset standard time has elapsed since receiving the artist message, the control unit (141) provides the fan message and private message stored in the individual database (1320) to the fan terminal (200).

[0190] Below, referring to Figure 14, a method for updating a private chat when an artist requests a private chat inquiry is described in detail.

[0191] Figure 14 is a flowchart of a private chat update method of an artist terminal according to an embodiment of the present invention.

[0192] Referring to FIG. 14, the private chat update method of the artist terminal (200) includes a private chat update request reception step (S410), an artist transmission database query step (S420), an artist reception database (S430), and a private chat update step (S440).

[0193] When an artist requests to view a private chat in the private chat update request reception step (S410), the control unit (141) receives a private chat inquiry request from the artist terminal (200-1). At this time, the private chat inquiry request includes DM ID information.

[0194] In the artist transmission database query step (S420), the control unit (141) requests the artist transmission database (1322) to search for a message corresponding to the DM ID. Accordingly, the artist transmission database (1322) responds with a list of artist message IDs stored corresponding to the artist's DM ID.

[0195] In the artist reception database query step (S430), the control unit (141) requests the artist reception database (1323) to search for a message corresponding to the DM ID. Accordingly, the artist reception database (1323) responds with a list of fan message IDs stored corresponding to the artist's DM ID.

[0196] Accordingly, the control unit (141) receives an artist message ID list from the artist transmission database (1322) and a fan message ID list from the artist reception database (1323). The control unit (141) requests a message information inquiry from the message body storage unit (1332) based on the received artist message ID list and fan message ID list.

[0197] Accordingly, the message body storage unit (1332) responds to the control unit (141) with message information corresponding to the artist message ID list and the fan message ID list. Then, in the private chat update step (S440), the control unit (141) that has received the artist message and fan message information updates the message content to the artist terminal (200).

[0198] Hereinafter, the artist message and fan message filtering method will be described in detail with reference to FIGS. 15 to 19.

[0199] Figure 15 is a conceptual diagram illustrating the function of a processor according to an embodiment of the present invention.

[0200] Referring to FIG. 15, the processor (140) may include, in addition to the functions of the control unit (141) described above, a taboo word filtering module (142) for filtering artist messages and fan messages, and a malicious message filtering module (143) for filtering malicious messages.

[0201] The forbidden word filtering module (142) determines whether artist messages and fan messages contain preset forbidden words. The forbidden words can be individually set for each artist and fan. Furthermore, for fans, the forbidden words can be set differently depending on the artist receiving the message.

[0202] The taboo word filtering module (142) identifies words included in artist messages and fan messages and determines whether the identified words correspond to preset taboo words.

[0203] In addition, the swear word filtering module (142) can convert fan messages and artist message data into image data. Then, the swear word filtering module (142) derives text corresponding to the converted image data.

[0204] The swear word filtering module (142) identifies words included in the text derived from the image, determines whether the identified words correspond to swear words, and can perform swear word filtering on artist messages and fan messages.

[0205] The swear word filtering module (142) uses a CNN (Convolution Neural Network) learning model to convert artist message data and fan message data into image data.

[0206] Specifically, the swear word filtering module (142) can convert into consonants and vowels of Hangul respectively when deriving text corresponding to the image data. For example, when the artist message or fan message contains content such as "んЙ刀│○ㅑ", '人ㅣ凹卜', etc., the swear word filtering module (142) converts the corresponding message into an image and derives consonants "ㅅ, ㄲ, ㅇ" and "ㅅ, ㅂ", and vowels "ㅐ, ㅣ, ㅑ" and "ㅣ, ㅏ". Then, it derives combinations of vowels and consonants, such as "새끼야" and "시바", as text.

[0207] Furthermore, CNNs can be composed of a feature extraction section that effectively recognizes and emphasizes features from adjacent images while maintaining the spatial information of the image, and a classifier section that classifies the image. The feature extraction section consists of a convolutional layer that finds image features while minimizing the number of shared parameters using filters, and a pooling layer that enhances and aggregates the features. Since CNNs can be retrained based on existing networks for new recognition tasks, CNN learning models can filter out new malicious messages that have been creatively transformed to avoid forbidden words.

[0208] Additionally, the taboo word filtering module (142) can filter whether the text derived using the Bidirectional Encoder Representation of Transformer (BERT) learning model contains taboo words. The BERT learning model can be fine-tuned using the accuracy of the judgment results.

[0209] BERT is a pre-trained model used by Google to label and train on vast amounts of text on the web using semi-supervised learning. BERT is used in the embedding process for tasks such as named entity recognition and text classification.

[0210] As a method for determining whether a forbidden word is included in a message by the forbidden word filtering module (142), a method using an algorithm that labels a message containing a vocabulary set by the user or a method using a judgment based on preset criteria can be used.

[0211] Pre-established standards can be categorized into expressions that include general profanity, vulgar and degrading expressions that are offensive to others, sexually provocative expressions, expressions that involve physical threats, expressions of discrimination based on region / race / country / religion, etc., and derogatory expressions that cause humiliation and shame to others.

[0212] The forbidden word filtering module (142) can filter messages by outputting '0' as a value for messages containing profanity or slanderous expressions, and outputting '1' as a value for messages not containing profanity or slanderous expressions.

[0213] The context of the text derived from the malicious message filtering module (143) and the forbidden word filtering module (142) is identified, and based on the identified context, the fan message is filtered to determine whether it corresponds to a malicious message.

[0214] The malicious message filtering module (143) classifies the context of a fan message by performing encoding using bidirectional context analysis on the fan message. More specifically, the malicious message filtering module (143) can filter out whether the fan message is malicious by inputting the fan message into a BERT (Bidirectional Encoder Representation of Transformer) learning model. The BERT learning model used in the malicious message filtering module (143) receives a single embedding that combines token embedding, segment embedding, and position embedding for the fan message and identifies the context of the fan message.

[0215] In addition, the malicious message filtering module (143) prevents out-of-vocabulary of words that make up fan messages by using word piece embedding that segments words into sub-word units as token embedding.

[0216] Specifically, the malicious message filtering module (143) can process token embedding using the word-piece embedding method. In the case of word-piece embedding, embedding can be performed in units smaller than words, and the longest subword can be set as a single unit. That is, frequently appearing subwords can be utilized as units for embedding themselves, and infrequently appearing words (rare words) can be separated into subwords. The existing word embedding method has a vocabulary outflow problem, making it difficult to learn or translate rare words, names, numbers, or words that do not exist in the vocabulary, whereas the word-piece embedding method can be applied to all languages, and since it segments words into subword units, it is effective in solving the vocabulary outflow problem while also increasing accuracy.

[0217] In addition, the malicious message filtering module (143) can input two sentences along with a sentence delimiter ([SEP]). At this time, due to the limitation on the input length, the total length of the two sentences may be limited to 512 subwords or less. In other words, since the learning time increases squarely as the input length increases, the input length must be set appropriately. Since the Korean language consists of an average of 20 subwords and 99% of the sentences do not exceed 60 subwords, the input length can be limited to 128 by combining the two sentences. However, since there may be long sentences, a method may be used in which the input length is first limited to 128 and learning is performed, and additional learning is performed in the final stage for inputs longer than 128.

[0218] Additionally, the malicious message filtering module (143) may use position encoding. For this purpose, a Transformer model may be applied, and the Transformer model may use a Self-Attention model instead of a CNN or RNN model. In this case, since the Self-Attention model does not reflect the position of the input, additional position information regarding the input token must be input. Therefore, Positional encoding using the Sinusoid function was used in the Transformer model, and this can be modified to use positional encoding in the malicious message filtering module (143). In positional encoding, encoding can be performed simply in the order of tokens, such as 0, 1, 2, ...

[0219] Additionally, the malicious message filtering module (143) can combine token embeddings, segment embeddings, and position embeddings to generate a single embedding (i.e., a single embedding). The single embedding can be generated in the form of a feature map, and can be transformed into an appropriate form for subsequent operation before use. For example, the malicious message filtering module (143) can use the results of applying layer normalization and dropout to the single embedding as input.

[0220] In addition, the malicious message filtering module (143) can increase the accuracy of malicious message judgment by providing the output malicious message encoding value to a fine tuning network. That is, the malicious message filtering module (143) can receive two sentences regarding a user's message as input along with a sentence delimiter ([SEP]), and can be constructed through pre-learning to generate a malicious message encoding value regarding the possibility of a malicious message as output through an internal inference step.

[0221] The malicious message filtering module (143) can perform malicious comment filtering operations by utilizing the BERT learning model as is, but the accuracy of malicious message identification can be improved through additional training. To this end, the malicious message filtering module (143) defines a model architecture that combines the BERT learning model with a fine-tuned network, which can be utilized in the malicious message identification process.

[0222] The malicious message filtering module (143) can determine the possibility of a fan message being a malicious message by providing the malicious message encoding value to a one-dimensional convolution layer (1D Convolution Layer) or by performing an ensemble operation based on a variable kernel size to generate a feature map and providing the feature map to a BiLSTM (Bidirectional Long Short-Term Memory) network.

[0223] In addition, the malicious message filtering module (143) can perform binary classification on the possibility of malicious messages by passing the results output through the BiLSTM network through the affine and softmax layers. Here, BiLSTM (Bidirectional Long Short-Term Memory) can correspond to a structure that uses two independent LSTM architectures together. First, the bidirectional LSTM can sequentially input sentences.

[0224] That is, it can receive sentences sequentially from left to right, just like a human. Furthermore, bidirectional LSTMs can be used in conjunction with reverse LSTMs, which read sentences from right to left, to consider the context behind the sentence. When making predictions in the output layer, bidirectional LSTMs can utilize both types of information by concatenating the outputs of forward and reverse LSTMs.

[0225] These bidirectional LSTMs enable end-to-end learning, simultaneously learning all parameters while minimizing loss in output values. They can also improve performance by internalizing word-phrase similarities into input vectors. Furthermore, bidirectional LSTMs can benefit from the inherent performance of LSTMs and the introduction of an attention mechanism, which allows for performance to remain intact even with long data lengths.

[0226] In addition, the malicious message filtering module (143) can implement a malicious message filter by combining a BERT learning model and a CNN (Convolutional Neural Network) learning model. More specifically, the malicious message filtering module (143) can input a malicious message encoding value to a one-dimensional convolutional layer (1D Convolution Layer), input the output of the one-dimensional convolutional layer to a GeLU layer, input the output of the GeLU layer to a MaxPooling layer, and generate a binary classification result based on the output of the MaxPooling layer. At this time, a linear classification method can be applied to the binary classification.

[0227] For example, linear classification methods may include linear regression or linear classification models.

[0228] In addition, the malicious message filtering module (143) can implement a malicious message filter by combining a BERT learning model and an ensemble CNN model. More specifically, the malicious message filtering module (143) can perform a first step of performing an ensemble operation based on a variable kernel size based on a malicious message encoding value, a second step of independently inputting the results of the ensemble operation to a GeLU layer, a third step of independently inputting the outputs of the GeLU layer to a MaxPooling layer, a fourth step of sequentially connecting the outputs of the MaxPooling layer to generate an intermediate value, a fifth step of inputting the intermediate value to an affine layer, a sixth step of inputting the output of the affine layer to a softmax layer, and a seventh step of generating a binary classification result based on the result of the softmax layer.

[0229] Here, the ensemble operation based on variable kernel sizes may correspond to multiple convolution operations performed by applying kernel sizes differently to the same malicious message encoding value. In addition, the convolution operation may include a 1D convolution operation. That is, the malicious comment judgment unit (230) may independently apply kernels with different sizes to the feature map corresponding to the malicious comment encoding value, thereby performing multiple convolution operations. Thereafter, the malicious message filtering module (143) may perform a classification operation regarding the possibility of a fan message being a malicious message through a step of integrating the results of each convolution operation into one (i.e., the fourth step).

[0230] In addition, the malicious message filtering module (143) can implement a malicious message filter by combining a BERT learning model, an ensemble CNN model, and a BiLSTM. More specifically, the malicious message filtering module (143) can perform a first step of performing an ensemble operation based on a variable kernel size based on a malicious comment encoding value, a second step of independently inputting the results of the ensemble operation to a GeLU layer, a third step of independently inputting the outputs of the GeLU layer to a MaxPooling layer, a fourth step of sequentially connecting the outputs of the MaxPooling layer to generate an intermediate value, a fifth step of applying the intermediate value to a BiLSTM (Bidirectional LSTM) (Long Short-Term Memory) model, a sixth step of inputting the result of the application to the BiLSTM model to an affine layer, a seventh step of inputting the output of the affine layer to a softmax layer, and an eighth step of generating a binary classification result based on the result of the softmax layer.

[0231] That is, the malicious message filtering module (143) can implement a malicious message filter by adding a step of applying an intermediate value to a BiLSTM (Bidirectional LSTM) (Long Short-Term Memory) model between the fourth and fifth steps of the method of implementing a malicious message filter by combining a BERT learning model and an ensemble CNN model.

[0232] In addition, for filtering malicious messages, the control unit (141) may derive a malicious message prediction score based on the frequency of fan use of forbidden words, contextual information of fan messages, and the frequency of fan writing malicious messages. If the derived malicious message prediction score is equal to or higher than a preset first reference score, the control unit (141) may provide a fan message review request notification to the fan terminal (200). In addition, if the derived malicious message prediction score is equal to or higher than a preset second reference score, the control unit (141) may provide a notification to stop sending fan messages. In this case, the second reference score is set to have a higher value than the first reference score.

[0233] Hereinafter, the data processing process for filtering fan messages is described in detail with reference to FIG. 16.

[0234] Figure 16 is a data flow diagram for fan message filtering according to an embodiment of the present invention.

[0235] Referring to FIG. 16, for fan messages, a forbidden word filtering step and a malicious message filtering step may be additionally performed for message filtering.

[0236] Specifically, when the control unit (141) receives a fan message from the fan terminal (200), the control unit (141) requests the forbidden word filtering module (142) to filter forbidden words for the fan message.

[0237] The forbidden word filtering module (142) that receives the forbidden word filtering request filters whether the fan message contains forbidden words based on the fan forbidden words set for the receiving artist.

[0238] Additionally, as described with reference to FIG. 15, the forbidden word filtering module (142) can convert a fan message into an image. The forbidden word filtering module (142) derives text corresponding to the converted image data, identifies words included in the derived text, and determines whether the identified words correspond to forbidden words, thereby performing forbidden word filtering of the fan message.

[0239] The forbidden word filtering module (142) responds with the fan message filtering result to the control unit (141), and if the fan message does not contain a forbidden word, the control unit (141) performs a request to store the fan message to the message body storage unit (1332) and a request to store the fan message to an individual database (1320).

[0240] When the fan message is completely stored in the individual database (1320), the control unit (141) requests the malicious message filtering module (143) to filter the malicious message. The malicious message filtering module (143) that receives the request for malicious message filtering classifies the context of the fan message and filters whether the fan message corresponds to a malicious message based on the classified context.

[0241] As described with reference to FIG. 15, the malicious message filtering module (143) classifies the context of a fan message by performing encoding using bidirectional context analysis on the fan message. The malicious message filtering module (143) can filter out whether the fan message is malicious by inputting the fan message into a BERT (Bidirectional Encoder Representation of Transformer) learning model. The BERT learning model used in the malicious message filtering module (143) receives a single embedding that combines token embedding, segment embedding, and position embedding for the fan message, identifies the context of the fan message, and classifies the malicious message.

[0242] The control unit (141), which receives the malicious message filtering result from the malicious message filtering module (143), derives a malicious message prediction score based on the frequency of use of forbidden words by the received fan, the context information of the fan message, and the frequency of writing malicious messages by the fan.

[0243] Additionally, the control unit (141) may provide a fan message review request notification to the fan terminal (200) if the derived malicious message prediction score is equal to or greater than a preset first reference score. Additionally, the control unit (141) may provide a fan message transmission suspension notification if the derived malicious message prediction score is equal to or greater than a preset second reference score. In this case, the second reference score is set to have a higher value than the first reference score.

[0244] If the result of filtering malicious messages indicates that the fan message is not a malicious message, the control unit (141) performs a subscription data lookup using the subscription information storage unit (1332) and requests the fan message to be stored in the artist reception database (1323).

[0245] Hereinafter, with reference to FIG. 17, the data processing process for filtering artist messages is described.

[0246] Figure 17 is a data flow diagram for artist message filtering according to an embodiment of the present invention.

[0247] Referring to Figure 17, for artist messages, an additional step of filtering forbidden words can be performed for message filtering.

[0248] Specifically, when the control unit (141) receives an artist message from the artist terminal (200), the control unit (141) requests the forbidden word filtering module (142) to filter forbidden words for the artist message.

[0249] The forbidden word filtering module (142) that receives the forbidden word filtering request filters whether the artist message contains a forbidden word based on the artist forbidden words set for the artist who wrote the message.

[0250] Additionally, as described with reference to FIG. 15, the taboo word filtering module (142) can convert artist messages into images. The taboo word filtering module (142) derives text corresponding to the converted image data, identifies words included in the derived text, and determines whether the identified words correspond to taboo words, thereby performing taboo word filtering of the artist message.

[0251] The forbidden word filtering module (142) responds with the artist message filtering result to the control unit (141), and if the artist message does not contain a forbidden word, the control unit (141) performs a request to store the artist message in the message body storage unit (1332), a request to store the artist message in the artist transmission database (1322), a request to store the artist message in the DM-specific database (1310), and a request to store the private message in the individual database (1320).

[0252] Hereinafter, a method for filtering fan messages will be described in detail with reference to FIG. 18.

[0253] Figure 18 is a flowchart of a fan message filtering method according to an embodiment of the present invention.

[0254] Referring to FIG. 18, in the case of the fan message filtering method, a forbidden word filtering step (S510) and a malicious message filtering step (S520) may be additionally included in the fan message storage method illustrated in FIG. 10.

[0255] In the forbidden word filtering step (S510), when the control unit (141) receives a fan message from the fan terminal (200), the control unit (141) requests the forbidden word filtering module (142) to filter the fan message for forbidden words.

[0256] In the forbidden word filtering step (S510), the forbidden word filtering module (142) that has received the forbidden word filtering request filters whether the fan message contains a forbidden word based on the fan forbidden words set for the receiving artist.

[0257] In the forbidden word filtering step (S510), the forbidden word filtering module (142) can convert a fan message into an image. The forbidden word filtering module (142) derives text corresponding to the converted image data, identifies words included in the derived text, and determines whether the identified words correspond to forbidden words, thereby performing forbidden word filtering of the fan message.

[0258] The forbidden word filtering module (142) responds with the fan message filtering result to the control unit (141), and the control unit (141) performs the fan message storage step (S120) in an individual database if the fan message does not contain a forbidden word.

[0259] In the swear word filtering step (S510), the swear word filtering module (142) converts artist message data and fan message data into image data using a CNN (Convolution Neural Network) learning model.

[0260] Specifically, the swear word filtering module (142) can convert text corresponding to image data into consonants and vowels of Hangul respectively when deriving the text. For example, when the artist message or fan message contains content such as "んЙ刀│○ㅑ", '人ㅣ凹卜', etc., the swear word filtering module (142) converts the corresponding message into an image and derives the consonants "ㅅ, ㄲ, ㅇ" and "ㅅ, ㅂ", and the vowels "ㅐ, ㅣ, ㅑ" and "ㅣ, ㅏ". Then, combinations of vowels and consonants, such as "새끼야" and "시바", are derived as text.

[0261] In addition, CNN can be composed of a part that extracts the features of an image in a way that effectively recognizes and emphasizes the features with adjacent images while maintaining the spatial information of the image, and a part that classifies the image. The feature extraction area is composed of a convolution layer that uses a filter to find the features of the image while minimizing the number of shared parameters, and a pooling layer that enhances and collects the features. Since it is possible to re - learn and use CNN for new recognition tasks based on the existing network, the CNN learning model can filter new malicious messages creatively modified to avoid swear words.

[0262] Additionally, in the forbidden word filtering step (S510), the forbidden word filtering module (142) can filter whether the text derived using a Bidirectional Encoder Representation of Transformer (BERT) learning model contains forbidden words. The BERT learning model can be fine-tuned using the accuracy of the judgment results.

[0263] BERT is a pre-trained model used by Google to label and train on vast amounts of text on the web using semi-supervised learning. BERT is used in the embedding process for tasks such as named entity recognition and text classification.

[0264] As a method for determining whether a forbidden word is included in a message by the forbidden word filtering module (142), a method using an algorithm that labels a message containing a vocabulary set by the user or a method using a judgment based on preset criteria can be used.

[0265] Pre-established standards can be categorized into expressions that include general profanity, vulgar and degrading expressions that are offensive to others, sexually provocative expressions, expressions that involve physical threats, expressions of discrimination based on region / race / country / religion, etc., and derogatory expressions that cause humiliation and shame to others.

[0266] In the forbidden word filtering step (S510), the forbidden word filtering module (142) can filter messages by outputting '0' as a value in the case of a message containing abusive or slanderous expressions, and outputting '1' as a value in the case of a message not containing abusive or slanderous expressions.

[0267] When the fan message is completely stored in the individual database (1320), in the malicious message filtering step (S520), the control unit (141) requests the malicious message filtering module (143) to filter the malicious message. In the malicious message filtering step (S520), the malicious message filtering module (143) that has received the malicious message filtering request classifies the context of the fan message and filters whether the fan message corresponds to a malicious message based on the classified context.

[0268] In the malicious message filtering step (S520), the context of the text derived from the malicious message filtering module (143) and the forbidden word filtering module (142) is identified, and based on the identified context, whether the fan message corresponds to a malicious message is filtered.

[0269] To this end, the malicious message filtering module (143) performs encoding using bidirectional context analysis on fan messages to classify the context of the fan messages.

[0270] In the malicious message filtering step (S520), the malicious message filtering module (143) can filter out whether the fan message is malicious by inputting the fan message into a BERT (Bidirectional Encoder Representation of Transformer) learning model. The BERT learning model used in the malicious message filtering module (143) receives a single embedding that combines token embedding, segment embedding, and position embedding for the fan message, identifies the context of the fan message, and classifies the malicious message.

[0271] In the malicious message filtering step (S520), the malicious message filtering module (143) classifies the context of the fan message by performing encoding using bidirectional context analysis on the fan message. More specifically, the malicious message filtering module (143) can filter out whether the fan message is malicious by inputting the fan message into a BERT (Bidirectional Encoder Representation of Transformer) learning model. The BERT learning model used in the malicious message filtering module (143) receives a single embedding that combines token embedding, segment embedding, and position embedding for the fan message and identifies the context of the fan message.

[0272] In addition, the malicious message filtering module (143) prevents out-of-vocabulary of words that make up fan messages by using word piece embedding that segments words into sub-word units as token embedding.

[0273] Specifically, the malicious message filtering module (143) can process token embedding using the word-piece embedding method. In the case of word-piece embedding, embedding can be performed in units smaller than words, and the longest subword can be set as a single unit. That is, frequently appearing subwords can be utilized as units for embedding themselves, and infrequently appearing words (rare words) can be separated into subwords. The existing word embedding method has a vocabulary outflow problem, making it difficult to learn or translate rare words, names, numbers, or words that do not exist in the vocabulary, whereas the word-piece embedding method can be applied to all languages, and since it segments words into subword units, it is effective in solving the vocabulary outflow problem while also increasing accuracy.

[0274] In addition, in the malicious message filtering step (S520), the malicious message filtering module (143) can input two sentences together with a sentence delimiter ([SEP]). At this time, due to the limitation on the input length, the total length of the two sentences may be limited to 512 subwords or less. In other words, since the learning time increases squarely as the length of the input increases, the length of the input must be set appropriately. Since the Korean language consists of an average of 20 subwords and 99% of the sentences do not exceed 60 subwords, the length of the input can be limited to 128 by combining the two sentences. However, since there may be long sentences, a method may be used in which the length of the input is first limited to 128 and learning is performed, and additional learning is performed in the final step for inputs longer than 128.

[0275] In addition, in the malicious message filtering step (S520), the malicious message filtering module (143) may use position encoding. For this purpose, a Transformer model may be applied, and the Transformer model may use a Self-Attention model instead of a CNN or RNN model. In this case, since the Self-Attention model does not reflect the position of the input, additional position information regarding the input token must be input. Therefore, Positional encoding using the Sinusoid function was used in the Transformer model, and this can be modified to use positional encoding in the malicious message filtering module (143). In positional encoding, encoding can be performed simply in the order of tokens, such as 0, 1, 2, ...

[0276] Additionally, in the malicious message filtering step (S520), the malicious message filtering module (143) can combine token embedding, segment embedding, and position embedding to generate a single embedding (i.e., a single embedding) value. At this time, the single embedding can be generated in the form of a feature map, and can be used after being transformed into an appropriate form according to the subsequent operation process. For example, the malicious message filtering module (143) can use as input the result of applying layer normalization and dropout to the single embedding.

[0277] In addition, in the malicious message filtering step (S520), the malicious message filtering module (143) can increase the accuracy of malicious message judgment by providing the output malicious message encoding value to a fine tuning network. That is, the malicious message filtering module (143) can receive two sentences regarding the user's message as input along with a sentence delimiter ([SEP]), and can be constructed through pre-learning to generate a malicious message encoding value regarding the possibility of a malicious message as output through an internal inference step.

[0278] In the malicious message filtering step (S520), the malicious message filtering module (143) can basically perform the malicious comment filtering operation by utilizing the BERT learning model as is, but the accuracy of malicious message judgment can be improved through additional learning. To this end, the malicious message filtering module (143) can define a model architecture that combines the BERT learning model and a fine-tuned network and utilize it in the malicious message judgment process.

[0279] The malicious message filtering module (143) can determine the possibility of a fan message being a malicious message by providing the malicious message encoding value to a one-dimensional convolution layer (1D Convolution Layer) or by performing an ensemble operation based on a variable kernel size to generate a feature map and providing the feature map to a BiLSTM (Bidirectional Long Short-Term Memory) network.

[0280] And, in the malicious message filtering step (S520), the malicious message filtering module (143) can perform binary classification on the possibility of malicious messages by passing the result output through the BiLSTM network to the Affine and Softmax layers. Here, BiLSTM (Bidirectional Long Short-Term Memory) and bidirectional LSTM can correspond to a structure that uses two independent LSTM architectures together. First, the bidirectional LSTM can sequentially input sentences.

[0281] That is, it can receive sentences sequentially from left to right, just like a human. Furthermore, bidirectional LSTMs can be used in conjunction with reverse LSTMs, which read sentences from right to left, to consider the context behind the sentence. When making predictions in the output layer, bidirectional LSTMs can utilize both types of information by concatenating the outputs of forward and reverse LSTMs.

[0282] These bidirectional LSTMs enable end-to-end learning, simultaneously learning all parameters while minimizing loss in output values. They can also improve performance by internalizing word-phrase similarities into input vectors. Furthermore, bidirectional LSTMs can benefit from the inherent performance of LSTMs and the introduction of an attention mechanism, which allows for performance to remain intact even with long data lengths.

[0283] In addition, in the malicious message filtering step (S520), the malicious message filtering module (143) can implement a malicious message filter by combining the BERT learning model and the CNN (Convolutional Neural Network) learning model. More specifically, the malicious message filtering module (143) can input the malicious message encoding value to a one-dimensional convolutional layer (1D Convolution Layer), input the output of the one-dimensional convolutional layer to a GeLU layer, input the output of the GeLU layer to a MaxPooling layer, and generate a binary classification result based on the output of the MaxPooling layer. At this time, a linear classification method can be applied to the binary classification.

[0284] For example, linear classification methods may include linear regression or linear classification models.

[0285] In addition, in the malicious message filtering step (S520), the malicious message filtering module (143) can implement a malicious message filter by combining the BERT learning model and the ensemble CNN model. More specifically, the malicious message filtering module (143) can perform a first step of performing an ensemble operation based on a variable kernel size based on a malicious message encoding value, a second step of independently inputting the results of the ensemble operation to a GeLU layer, a third step of independently inputting the outputs of the GeLU layer to a MaxPooling layer, a fourth step of sequentially connecting the outputs of the MaxPooling layer to generate an intermediate value, a fifth step of inputting the intermediate value to an affine layer, a sixth step of inputting the output of the affine layer to a softmax layer, and a seventh step of generating a binary classification result based on the result of the softmax layer.

[0286] Here, the ensemble operation based on variable kernel sizes may correspond to multiple convolution operations performed by applying kernel sizes differently to the same malicious message encoding value. In addition, the convolution operation may include a 1D convolution operation. That is, the malicious comment judgment unit (230) may independently apply kernels with different sizes to the feature map corresponding to the malicious comment encoding value, thereby performing multiple convolution operations. Thereafter, the malicious message filtering module (143) may perform a classification operation regarding the possibility of a fan message being a malicious message through a step of integrating the results of each convolution operation into one (i.e., the fourth step).

[0287] Additionally, in the malicious message filtering step (S520), the malicious message filtering module (143) can implement a malicious message filter by combining a BERT learning model, an ensemble CNN model, and BiLSTM. More specifically, the malicious message filtering module (143) can perform a first step of performing an ensemble operation based on a variable kernel size based on a malicious comment encoding value, a second step of independently inputting the results of the ensemble operation to a GeLU layer, a third step of independently inputting the outputs of the GeLU layer to a MaxPooling layer, a fourth step of sequentially connecting the outputs of the MaxPooling layer to generate an intermediate value, a fifth step of applying the intermediate value to a BiLSTM (Bidirectional LSTM) (Long Short-Term Memory) model, a sixth step of inputting the result of the application to the BiLSTM model to an affine layer, a seventh step of inputting the output of the affine layer to a softmax layer, and an eighth step of generating a binary classification result based on the result of the softmax layer.

[0288] That is, the malicious message filtering module (143) can implement a malicious message filter by adding a step of applying an intermediate value to a BiLSTM (Bidirectional LSTM) (Long Short-Term Memory) model between the fourth and fifth steps of the method of implementing a malicious message filter by combining a BERT learning model and an ensemble CNN model.

[0289] In addition, in the malicious message filtering step (S520), the control unit (141) can derive a malicious message prediction score based on the frequency of the fan's use of forbidden words, the context information of the fan message, and the frequency of the fan's writing of malicious messages for malicious message filtering.

[0290] In the malicious message filtering step (S520), the control unit (141) may provide a fan message review request notification to the fan terminal (200) if the derived malicious message prediction score is equal to or greater than a preset first reference score. Furthermore, the control unit (141) may provide a fan message transmission suspension notification if the derived malicious message prediction score is equal to or greater than a preset second reference score. In this case, the second reference score is set to have a higher value than the first reference score.

[0291] If the fan message is not a malicious message as a result of the malicious message filtering, the message feed provision step (S130) is performed.

[0292] Hereinafter, the artist message filtering method is described in detail with reference to FIG. 19.

[0293] Figure 19 is a flowchart of an artist message filtering method according to an embodiment of the present invention.

[0294] Referring to FIG. 19, in the case of the artist message filtering method, a taboo word filtering step (S511) can be additionally performed in the artist message storage method illustrated in FIG. 11.

[0295] In the forbidden word filtering step (S511), when the control unit (141) receives an artist message from the artist terminal (200), the control unit (141) requests the forbidden word filtering module (142) to filter forbidden words for the artist message.

[0296] In the forbidden word filtering step (S511), the forbidden word filtering module (142) that has received the forbidden word filtering request filters whether the artist message contains a forbidden word based on the artist forbidden words set for the artist who wrote the message.

[0297] As described above, the forbidden word filtering module (142) can convert an artist message into an image. The forbidden word filtering module (142) derives text corresponding to the converted image data, identifies words included in the derived text, and determines whether the identified words correspond to forbidden words, thereby performing forbidden word filtering of the artist message.

[0298] The forbidden word filtering module (142) responds with the artist message filtering result to the control unit (141), and the control unit (141) performs the artist message storage step (S220) in the individual database (1320), the artist message storage step (S230) in the DM-specific database (1310), the private message generation step (S240), and the private message storage step (S250) in the individual database (1321) when the artist message does not contain a forbidden word.

[0299] In the description of the present invention described above, the artist terminal (200-1) and the fan terminal (200-2) may correspond to an artist account and a fan account, or a terminal logged in using an artist account and a terminal logged in using a fan account.

[0300] The foregoing description of the present invention is for illustrative purposes only, and those skilled in the art will readily appreciate that the present invention can be readily modified into other specific forms without altering the technical spirit or essential characteristics of the present invention. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. For example, each component described as a single entity may be implemented in a distributed manner, and similarly, components described as distributed may be implemented in a combined manner.

[0301] The scope of the present invention is indicated by the claims described below rather than the detailed description above, and all changes or modifications derived from the meaning and scope of the claims and their equivalent concepts should be interpreted as being included in the scope of the present invention.

[0302] The form for carrying out the invention is the same as the best form for carrying out the invention described above.

[0303] The present invention relates to a method and device for providing a private chat service, and is applicable to the entertainment industry and messenger-related industries, and thus has industrial applicability.

Claims

1. A method for providing a private chat service by relaying messages between an artist terminal and multiple fan terminals on a private chat service relay server. Step of receiving an artist message from the artist terminal; A step of filtering whether the above artist message contains preset artist taboo words, If the artist message does not contain the artist's forbidden words, a step of storing the artist message and artist information in a database for each DM; A step of receiving a request for a message update in a private chat from a first fan terminal requesting to view the artist message among the plurality of fan terminals; A step of generating a private message corresponding to the first fan terminal based on the fan account information corresponding to the first fan terminal and the artist message, and A step of transmitting the above private message to the first fan terminal, A method for providing a private chat service, wherein the above DM database is configured to store messages by artist, and the above private message means a message in which specific words or keywords in the message are converted according to the recipient.

2. In paragraph 1, The step of filtering whether the above artist's forbidden words are included is: Step of converting the above artist message into an image, A step of deriving text based on the above image, and A step of filtering whether the artist's forbidden words are included based on the above text. A method for providing a private chat service, including:

3. In paragraph 2, The steps to convert the above artist message into an image are: A method for providing a private chat service, comprising a step of converting the artist message into an image using a CNN (Convolution Neural Network) learning model.

4. In paragraph 3, The step of filtering whether the above artist's forbidden words are included is: A method for providing a private chat service, comprising a step of filtering whether the text contains the artist's forbidden words using a BERT (Bidirectional Encoder Representation of Transformer) learning model.

5. In paragraph 4, The steps of saving the above artist messages and artist information in the DM database are as follows: A step of matching a plurality of individual databases corresponding to each of the plurality of fan terminals, A step of generating a plurality of private messages corresponding to each of the plurality of fan terminals based on the fan account information corresponding to each of the plurality of fan terminals and the artist message, and A step of storing the plurality of private messages in the individual database corresponding to the fan terminal based on the fan account information. A method for providing a private chat service, including:

6. In paragraph 5, The step of transmitting the above private message to the first fan terminal is: When the above multiple private messages are completely stored in the individual database, a step of deleting the artist messages stored in the DM-specific database, and A step of receiving a private message from a first individual database corresponding to a first fan terminal and transmitting the private message to the first fan terminal. A method for providing a private chat service, including:

7. In paragraph 6, The step of transmitting the above private message to the first fan terminal is: A step of deleting the artist messages stored in the DM-specific database after a preset standard time has elapsed since the artist messages and artist information are stored in the DM-specific database, and A step of receiving a private message from the first individual database and transmitting it to the first fan terminal. A method for providing a private chat service, including:

8. A method for providing a private chat service by relaying messages between an artist terminal and multiple fan terminals on a private chat service relay server. A step of matching a plurality of individual databases corresponding to each of the plurality of fan terminals, A step of receiving a fan message from a first fan terminal among multiple fan terminals, A step of filtering whether the above fan message contains preset fan taboo words, A step of storing the fan message in a first individual database corresponding to the first fan terminal among the plurality of individual databases; A step of classifying the context of the fan message using a learning-based artificial intelligence model; A step of classifying whether the fan message corresponds to a malicious message based on the classified context; A step of receiving an artist message from the artist terminal, Step of storing the above artist message and artist information in a DM-specific database; A step of receiving a message update request of a private chat from the first fan terminal; A step of generating a private message corresponding to the first fan terminal based on the fan account information corresponding to the first fan terminal and the artist message, and A step of transmitting the fan message and the private message stored in the first individual database to the first fan terminal, A method for providing a private chat service, wherein the above DM database is configured to store messages by artist, and the above private message means a message in which specific words or keywords in the message are converted according to the recipient.

9. In paragraph 8, The step of filtering whether the above fan taboo words are included is: Step of converting the above fan message into an image, A step of deriving text based on the above image, and A step of filtering whether the above text contains the above fan-banned words based on the above text. A method for providing a private chat service, including:

10. In paragraph 9, The step of converting the above fan message into an image is: A method for providing a private chat service, comprising a step of converting the fan message into an image using a CNN (Convolution Neural Network) learning model.

11. In paragraph 10, The step of filtering whether the above fan taboo words are included is: A method for providing a private chat service, comprising a step of filtering whether the text contains the fan-banned word using a BERT (Bidirectional Encoder Representation of Transformer) learning model.

12. In paragraph 9, The step of classifying the context of the above fan message is: A method for providing a private chat service, comprising a step of classifying the context of the fan message by performing encoding using a two-way context analysis on the fan message.

13. In paragraph 12, The step of classifying the context of the above fan message is: A step of inputting a single embedding that combines token embedding, segment embedding, and position embedding in the encoding process through the bidirectional context analysis into the BERT (Bidirectional Encoder Representation of Transformer) learning model. A method for providing a private chat service, further comprising:

14. In paragraph 11 or 13, The step of classifying whether the above fan message corresponds to a malicious message is as follows: A step of deriving a malicious message prediction score based on the frequency of the fan's use of forbidden words, the contextual information of the fan message, and the frequency of the artist's writing malicious messages. A step of providing a fan message review request notification when the malicious message prediction score is greater than or equal to a preset first reference score; A step of providing a notification to stop sending the fan message when the malicious message prediction score is higher than a preset second reference score. A method for providing a private chat service, including:

15. In paragraph 14, The step of transmitting the above private message to the first fan terminal is: When the above private message is completely stored in the individual database, a step of deleting the artist message stored in the DM-specific database, and A step of receiving the private message from the first individual database and transmitting it to the first fan terminal. A method for providing a private chat service, including:

16. In paragraph 15, The step of transmitting the above private message to the first fan terminal is: A step of deleting the artist messages stored in the DM-specific database after a preset standard time has elapsed since the artist messages and artist information are stored in the DM-specific database, and A step of receiving a private message from the first individual database and transmitting it to the first fan terminal. A method for providing a private chat service, including:

17. A device that provides a private chat service by relaying messages between artist terminals and multiple fan terminals on a private chat service relay server. A communication module that performs information transmission and reception between the terminal and the private chat service relay server; Memory to store private chat programs, A DM-specific database that stores messages by artist, Multiple individual databases allocated to each user and storing user-specific messages, and A processor for executing a private chat program stored in the above memory, The processor receives a fan message from a first fan terminal among a plurality of fan terminals, stores the fan message in a first individual database corresponding to the first fan terminal among the plurality of individual databases, receives an artist message from the artist terminal, filters whether the fan message and the artist message include preset fan taboo words and artist taboo words, and if the fan message does not include the fan taboo words, stores the artist message and artist information in a DM-specific database, and if a message update request of a private chat is received from the first fan terminal, generates a private message corresponding to the first fan terminal based on fan account information corresponding to the first fan terminal and the artist message, and transmits the fan message and the private message stored in the first individual database to the first fan terminal. The above private message is a private chat service providing device that means a message in which specific words or keywords in the message are converted according to the recipient.

Images