KR20220116719A - An apparatus for collecting and providing product sales information posted on a mobile messenger and a method therefor - Google Patents

Abstract

An apparatus for collecting and providing product sales information posted on a mobile messenger of the present invention includes: a collection unit connecting to a messenger server and collecting a plurality of pieces of product sales information from at least one between an open chat room and a channel; a selection unit selecting a plurality of pieces of product sales information having reliability higher than a predetermined level from the plurality of pieces of product sales information; and a recommendation unit organizing a recommendation list by arranging the plurality of pieces of selected product sales information according to a request of a user device and providing the organized recommendation list to the user device.

Description

An apparatus for collecting and providing product sales information posted on a mobile messenger and a method therefor

The present invention relates to a technology for collecting and providing product sales information, and more particularly, to an apparatus for collecting and providing product sales information posted on a mobile messenger, and a method therefor.

Mobile messengers, such as Telegram, provide the ability to create open chat rooms for free and participate without conditions. Many public chat rooms have been opened and used for the purpose of providing information for product sales. In particular, when opening a chat room by setting the title of a specific product group to sale, various sellers and buyers of the product group participate and become active in many cases. Since it is difficult to restrict users' posts due to the nature of the messenger, anyone can easily send a sales message to all users participating in the chat room, unless the chat room creator manually sanctions it.

A chat room in which only the creator can post messages among public chat rooms is called a channel. Any messenger user can participate as long as they know the channel address without special permission. When sellers open a channel, they can send the desired message to channel participants in detail in the channel they own. In order to activate the channel, users must inform the outside of their channel address so that they can participate. Usually, a product sales message including the channel address is posted on the public Internet or in an open product sales chat room to induce users to participate in the channel they own. In popular open chat rooms, many channel owners send product sales information and channel promotion messages.

In order for product buyers to obtain and compare product sales information such as sellers, sales prices, and transaction conditions posted in Telegram open chat rooms or channels, they must first use Telegram. There is the inconvenience of extracting, organizing, and comparing messages.

Korea Patent Publication No. 2013-0038534 (published on April 18, 2013)

An object of the present invention is to provide an apparatus and method for securing and providing product information posted in a public chat room and channel of a messenger.

According to a preferred embodiment of the present invention for achieving the above object, an apparatus for collecting and providing product sales information posted on a mobile messenger is a plurality of A collection unit that collects product sales information, a selection unit that selects a plurality of product sales information having a reliability higher than a predetermined level from the plurality of product sales information, and sorts and recommends the selected plurality of product sales information according to the request of the user device and a recommendation unit that configures the list and provides the configured recommendation list to the user device.

The collection unit detects links, images, and texts in the public chat room and the channels, and an image in which the area of the link is divided and displayed in the image, and the text to which the link is applied and the text to which the link is not applied in the text. Analysis data including the separated text may be generated, and the analysis data may be input to a detection model. Then, the detection model calculates a first detection vector and a second detection vector by performing a plurality of operations in which a weight between a plurality of layers is applied to the analysis data, and the first detection vector and the second detection vector A merge detection vector is derived by merging, and an output value representing the probability of whether product sales information is included in a public chat room or channel is calculated from the merge detection vector. Then, the collection unit is characterized in that it determines whether product sales information is included in the corresponding open chat room or channel according to the probability indicated by the calculated output value.

을 통해 클러스터의 중심과 복수의 상품판매정보벡터와의 차이를 나타내는 손실을 산출하고, 복수의 클러스터 각각에 대해 클러스터의 중심과 복수의 상품판매정보벡터와의 차이를 나타내는 손실의 평균 및 표준 편차로부터 수학식

에 따라 임계손실을 산출하고, 복수의 클러스터 각각에서 산출된 임계손실 이하의 손실을 가지는 상품판매정보벡터를 추출하고, 추출된 상품판매정보벡터에 대응하는 판매자를 신뢰도가 소정 수준 이상인 판매자로 선별하는 것을 특징으로 한다. 여기서, 상기 n은 인덱스이고, 상기 E는 손실을 나타내며, 상기 k는 클러스터의 인덱스이고, 상기 Mk는 k번째 클러스터의 중심을 나타내며, 상기 n은 상품판매정보벡터의 인덱스이고, 상기 Sn은 n번째 상품판매정보벡터이고, 상기 T는 임계손실이고, 상기 w는 클러스터의 중심과 복수의 상품판매정보벡터와의 차이를 나타내는 손실의 평균이고, 상기 σ는 클러스터의 중심과 복수의 상품판매정보벡터와의 차이를 나타내는 손실의 표준 편차이고, 상기

는 표준편차에 대한 가중치인 것을 특징으로 한다. The sorting unit embeds a plurality of product sales information in a predetermined vector space to generate a plurality of product sales information vectors, clusters the plurality of product sales information vectors into a plurality of clusters, and generates a convex hull in each of the clustered plurality of clusters. and set the center of gravity of the polygon constituting the convex hull in each of the plurality of clusters as the center of the cluster, and for each of the plurality of clusters, the equation

The loss representing the difference between the center of the cluster and the plurality of product sales information vectors is calculated through formula

Calculates the critical loss according to characterized in that Here, n is an index, E is a loss, k is an index of a cluster, Mk is a center of a k-th cluster, n is an index of a product sales information vector, and Sn is an n-th a product sales information vector, where T is a critical loss, w is an average of losses representing the difference between the center of a cluster and a plurality of product sales information vectors, and σ is the center of a cluster and a plurality of product sales information vectors is the standard deviation of the loss representing the difference in

is a weight for the standard deviation.

In a method for collecting and providing product sales information posted on a mobile messenger according to a preferred embodiment of the present invention for achieving the above object, the collection unit connects to the messenger server and opens at least one of open chat rooms and channels. The steps of collecting a plurality of product sales information, the selection unit selecting a plurality of product sales information having a reliability level higher than or equal to a predetermined level from the plurality of product sales information, and the information providing unit requesting the selected plurality of product sales information from the user device and composing a list of product sales information by arranging the information according to the method, and providing the configured list of product sales information to the user device.

The collecting of the plurality of product sales information includes: detecting, by the collecting unit, links, images, and texts in the open chat room and the channel; Generating analysis data including text in which the text to which the link is applied and the text to which the link is not applied are separated from text, the collecting unit inputting the analysis data into a detection model; A first detection vector and a second detection vector are calculated by performing a plurality of operations in which a weight between a plurality of layers is applied on the analysis data, and a merge detection vector is derived by merging the first detection vector and the second detection vector and calculating, from the merge detection vector, an output value indicating the probability of whether product sales information is included in the public chat room or channel, and product sales information in the public chat room or channel according to the probability indicated by the output value calculated by the collection unit and determining whether the product sales information is included, and collecting product sales information from an open chat room or channel determined to contain product sales information.

을 통해 클러스터의 중심과 복수의 상품판매정보벡터와의 차이를 나타내는 손실을 산출하는 단계와, 상기 선별부가 복수의 클러스터 각각에 대해 클러스터의 중심과 복수의 상품판매정보벡터와의 차이를 나타내는 손실의 평균 및 표준 편차로부터 수학식

에 따라 임계손실을 산출하는 단계와, 상기 선별부가 복수의 클러스터 각각에서 산출된 임계손실 이하의 손실을 가지는 상품판매정보벡터를 추출하고, 추출된 상품판매정보벡터에 대응하는 판매자를 신뢰도가 소정 수준 이상인 판매자로 선별하는 단계를 포함한다. 여기서, 상기 n은 인덱스이고, 상기 E는 손실을 나타내며, 상기 k는 클러스터의 인덱스이고, 상기 Mk는 k번째 클러스터의 중심을 나타내며, 상기 n은 상품판매정보벡터의 인덱스이고, 상기 Sn은 n번째 상품판매정보벡터이고, 상기 T는 임계손실이고, 상기 w는 클러스터의 중심과 복수의 상품판매정보벡터와의 차이를 나타내는 손실의 평균이고, 상기 σ는 클러스터의 중심과 복수의 상품판매정보벡터와의 차이를 나타내는 손실의 표준 편차이고, 상기

는 표준편차에 대한 가중치인 것을 특징으로 한다. The selecting of the plurality of product sales information having the reliability equal to or higher than a predetermined level includes: the selection unit embedding a plurality of product sales information in a predetermined vector space to generate a plurality of product sales information vectors; Clustering the sales information vector into a plurality of clusters, deriving a convex hull from each of the plurality of clustered clusters, and setting, by the selector, a center of gravity of a polygon constituting a convex hull in each of the plurality of clusters as the center of the cluster And, the selector for each of the plurality of clusters

calculating a loss representing the difference between the center of the cluster and the plurality of product sales information vectors through Equations from mean and standard deviation

calculating a critical loss according to Including the step of selecting the above sellers. Here, n is an index, E is a loss, k is an index of a cluster, Mk is a center of a k-th cluster, n is an index of a product sales information vector, and Sn is an n-th a product sales information vector, where T is a critical loss, w is an average of losses representing the difference between the center of a cluster and a plurality of product sales information vectors, and σ is the center of a cluster and a plurality of product sales information vectors is the standard deviation of the loss representing the difference in

is a weight for the standard deviation.

According to the present invention, by securing product information posted in the public chat room and channel of the mobile messenger, processing and providing it in a form desired by users, such as by product, by seller, by selling price, etc., products sold through mobile messenger conveniently by consumers can be purchased

1 is a diagram for explaining a system for collecting and providing product sales information posted on a mobile messenger according to an embodiment of the present invention.
2 and 3 are diagrams for explaining product sales information posted on a mobile messenger according to an embodiment of the present invention.
4 is a block diagram for explaining the configuration of a sales information server according to an embodiment of the present invention.
5 is a block diagram illustrating a detailed configuration of a sales information server according to an embodiment of the present invention.
6 is a block diagram illustrating the configuration of a detection model (DM) that is a deep learning model (DLM) according to an embodiment of the present invention.
7 to 9 are diagrams for explaining the detailed configuration of a detection model (DM) that is a deep learning model (DLM) according to an embodiment of the present invention.
10 is a block diagram for explaining the configuration of a user device according to an embodiment of the present invention.
11 is a flowchart illustrating a method for collecting and providing product sales information posted on a mobile messenger according to an embodiment of the present invention.
12 is a flowchart for explaining a deep learning method for a detection model that is a deep learning model according to an embodiment of the present invention.
13 is a flowchart illustrating a method of detecting a plurality of public chat rooms and channels including product sales information through a detection model according to an embodiment of the present invention.
14 is a flowchart for explaining a method of selecting a plurality of product sales information having a reliability equal to or higher than a predetermined level from a plurality of product sales information according to an embodiment of the present invention.
15 is a diagram for explaining a method of embedding and clustering a plurality of product sales information in a vector space according to an embodiment of the present invention.
FIG. 16 is a diagram for explaining a method of selecting product sales information having a reliability higher than or equal to a predetermined level from a plurality of product sales information according to an embodiment of the present invention.

In order to clarify the characteristics and advantages of the problem solving means of the present invention, the present invention will be described in more detail with reference to specific embodiments of the present invention shown in the accompanying drawings. However, detailed descriptions of well-known functions or configurations that may obscure the gist of the present invention in the following description and accompanying drawings will be omitted. In addition, it should be noted that throughout the drawings, the same components are denoted by the same reference numerals as much as possible.

The terms or words used in the following description and drawings should not be construed as being limited to conventional or dictionary meanings, and the inventor may properly define the concept of terms for describing his invention in the best way. Based on the principle that there is, it should be interpreted as meaning and concept consistent with the technical idea of the present invention. Therefore, the embodiments described in this specification and the configurations shown in the drawings are only the most preferred embodiment of the present invention, and do not represent all of the technical spirit of the present invention. It should be understood that there may be equivalents and variations.

In addition, terms including ordinal numbers such as 1st, 2nd, etc. are used to describe various components, and are used only for the purpose of distinguishing one component from other components, and to limit the components. not used For example, without departing from the scope of the present invention, the second component may be referred to as the first component, and similarly, the first component may also be referred to as the second component.

In addition, when an element is referred to as being “connected” or “connected” to another element, it means that it is logically or physically connected or can be connected. In other words, it should be understood that a component may be directly connected or connected to another component, but another component may exist in the middle, and may be indirectly connected or connected.

In addition, the terms used herein are used only to describe specific embodiments, and are not intended to limit the present invention. The singular expression includes the plural expression unless the context clearly dictates otherwise. In addition, terms such as "comprises" or "have" described in this specification are intended to designate that the features, numbers, steps, operations, components, parts, or combinations thereof described in the specification exist, but one or the It should be understood that the above does not preclude the possibility of addition or existence of other features or numbers, steps, operations, components, parts, or combinations thereof.

In addition, terms such as “…unit”, “…group”, and “module” described in the specification mean a unit that processes at least one function or operation, which may be implemented as hardware or software or a combination of hardware and software. have.

Also, "a or an", "one", "the" and similar terms are otherwise indicated herein in the context of describing the invention (especially in the context of the following claims). or may be used in a sense including both the singular and the plural unless clearly contradicted by context.

In addition, embodiments within the scope of the present invention include computer-readable media having or carrying computer-executable instructions or data structures stored thereon. Such computer readable media can be any available media that can be accessed by a general purpose or special purpose computer system. By way of example, such computer-readable media may be in the form of RAM, ROM, EPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage, or computer-executable instructions, computer-readable instructions, or data structures. It may include, but is not limited to, a physical storage medium such as any other medium that can be used to store or convey any program code means that has been created, and which can be accessed by a general purpose or special purpose computer system.

In the following description and claims, a “network” is defined as one or more data links that enable the transfer of electronic data between computer systems and/or modules. When information is transmitted or provided to a computer system over a network or other (wired, wireless, or a combination of wired or wireless) communication connection, the connection may be understood as a computer-readable medium. Computer readable instructions include, for example, instructions and data that cause a general purpose computer system or special purpose computer system to perform a particular function or group of functions. The computer executable instructions may be, for example, binary, intermediate format instructions such as assembly language, or even source code.

In addition, the present invention relates to personal computers, laptop computers, handheld devices, multiprocessor systems, microprocessor-based or programmable consumer electronics, network PCs, minicomputers, mainframe computers, mobile telephones, PDAs, pagers. In a network computing environment having various types of computer system configurations including (pager) and the like, it may be applied to the computer systems. The invention may also be practiced in distributed system environments where both local and remote computer systems linked through a network by a wired data link, a wireless data link, or a combination of wired and wireless data links perform tasks. In a distributed system environment, program modules may be located in local and remote memory storage devices.

First, a system for collecting and providing product sales information posted on a mobile messenger according to an embodiment of the present invention will be described. 1 is a diagram for explaining a system for collecting and providing product sales information posted on a mobile messenger according to an embodiment of the present invention. 2 and 3 are diagrams for explaining product sales information posted on a mobile messenger according to an embodiment of the present invention. Referring to FIG. 1 , a system for collecting and providing product sales information posted on a mobile messenger according to an embodiment of the present invention (hereinafter, abbreviated as 'sales information system') includes a sales information server 10 and a messenger server. (20) and a user device (30).

The sales information server 10 , the messenger server 20 , and the user device 30 may transmit and receive data to and from each other through the network NW. For example, the network NW may include wireless networks such as Wireless LAN (WLAN), Wi-Fi, Wibro, Wimax, High Speed Downlink Packet Access (HSDPA), etc., Wired networks such as Ethernet, xDSL (ADSL, VDSL), HFC (Hybrid Fiber Coaxial Cable), FTTC (Fiber to The Curb), and FTTH (Fiber To The Home) may be included depending on the system implementation method.

In addition, the network NW of the present invention may include, for example, a mobile communication network including a plurality of access networks and a core network connecting them. Here, the access network is a network that directly connects with the user device 30 to perform wireless communication, and includes, for example, a plurality of base stations such as BS (Base Station), BTS (Base Transceiver Station), NodeB, eNodeB, and the like, and BSC. (Base Station Controller), may be implemented as a base station controller such as RNC (Radio Network Controller). In addition, as described above, the digital signal processing unit and the radio signal processing unit integrated in the base station are respectively a digital unit (hereinafter referred to as 'DU') and a radio unit (hereinafter referred to as 'RU'). ), a plurality of RUs (not shown) may be installed in a plurality of areas, respectively, and the plurality of RUs may be connected to a centralized DU to be configured. In addition, the core network constituting the mobile network together with the access network serves to connect the access network and an external network, for example, other networks such as the Internet.

As described above, the core network is a network system that performs a main function for a mobile communication service such as mobility control and switching between access networks, and performs circuit switching or packet switching. Manages and controls packet flow within the network. In addition, the core network manages inter-frequency mobility, and may play a role for interworking with traffic in the access network and the core network and other networks, for example, the Internet. Such a core network may be configured to further include a Serving GateWay (SGW), a PDN GateWay (PGW), a Mobile Switching Center (MSC), a Home Location Register (HLR), a Mobile Mobility Entity (MME), and a Home Subscriber Server (HSS). may be

Also, the network NW according to the present invention may include the Internet. The Internet means an ordinary public network, that is, a public network, in which information is exchanged according to the TCP/IP protocol. The sales information server 10 , the messenger server 20 , and the user device 30 interwork with each other through this network NW to configure the sales information system according to the present invention.

The sales information server 10 basically collects product sales information by accessing at least one of the public chat rooms and channels running in the plurality of messenger servers 20 , and provides it to the user device 30 . For example, the sales information server 10 collects product sales information by accessing a telegram open chat room as shown in FIG. 2 or a telegram channel shown in FIG. can provide Here, the sales information server 10 is an entity existing on the network, and serves as a web server, a database server, and an application server.

The messenger server 20 is a server that provides a mobile messenger service, and provides a service that allows registered members to open and participate in chat rooms, public chat rooms, channels, and the like.

The user device 30 is a device used by a user who wants to purchase a product that can be purchased through an open chat room or channel of a mobile messenger. The user device 30 accesses the sales information server 10 , receives product sales information provided by the sales information server 10 , and displays it so that the user can read it. The user device 30 may be representatively exemplified by a smartphone, but is not limited thereto, and may be applied to various terminals such as all information and communication devices, multimedia terminals, wired terminals, fixed terminals, and Internet Protocol (IP) terminals. For example, the terminal is a mobile terminal having various mobile communication specifications such as a mobile phone, a portable multimedia player (PMP), a mobile Internet device (MID), a tablet PC, a phablet PC, and an information communication device, or a notebook computer, personal computer, etc. It may be a fixed terminal of

Next, the configuration of the sales information server 10 according to the embodiment of the present invention will be described in more detail. 4 is a block diagram for explaining the configuration of a sales information server according to an embodiment of the present invention. 5 is a block diagram illustrating a detailed configuration of a sales information server according to an embodiment of the present invention. Referring to FIG. 4 , the sales information server 10 includes a communication module 11 , a storage module 12 , and a control module 13 .

The communication module 11 is for communicating with the messenger server 20 and the user device 30 through a network. The communication module 11 may transmit/receive data to and from the messenger server 20 and the user device 30 . The communication module 11 may include a modem for modulating a signal to be transmitted and demodulating a received signal to transmit/receive data through a network. The communication module 11 may transmit data received from the control module 13 through the network. In addition, the communication module 11 may transmit data received through the network to the control module 13 .

The storage module 12 serves to store programs and data necessary for the operation of the sales information server 10 , and may be divided into a program area and a data area. The program area may store a program for controlling the overall operation of the sales information server 10 , an operating system (OS) for booting the sales information server 10 , an application for recommendation, and the like. The data area is an area in which data generated according to the operation of the sales information server 10 and data necessary for the operation of the sales information server 10 are stored. Each type of data stored in the storage module 120 may be deleted, changed, or added according to a user's operation.

The control module 13 may control the overall operation of the sales information server 10 and the signal flow between internal blocks of the sales information server 10 , and perform a data processing function of processing data. The control module 13 may be a central processing unit (CPU), a graphic processing unit (GPU), a digital signal processor (DSP), or the like.

Referring to FIG. 5 , the control module 13 includes a collection unit 200 , a selection unit 300 , and an information providing unit 300 .

The model generation unit 100 generates a detection model (DM), which is a deep learning model (DLM) according to an embodiment of the present invention, through deep learning, and generates the detection model (DM) It is provided to the collection unit 200 . Although the model generator 100 is described as being included in one device, it may be configured as a separate device.

The collection unit 200 connects to the messenger server 20 through the communication module 11 and detects a public chat room and channel including product sales information using a detection model (DM), and in the detected open chat room and channel This is to collect a plurality of product sales information. Here, the product sales information includes a product for sale, a seller, a sales volume, a sales price, a discount rate, a sales period, a recent sales date, a sales frequency, a link to a sales page, a link to a payment page, and the like.

The selection unit 300 selects product sales information having a reliability higher than or equal to a predetermined level from a plurality of product sales information, and newly adds or updates the selected plurality of product sales information to the storage module 12 and stores the selected product sales information.

The information providing unit 400 organizes a list of product sales information by arranging a plurality of stored product sales information, that is, product sales information having a reliability level higher than or equal to a predetermined level according to the request of the user device 30 , and uses the information in the communication module 11 . ) through the user device 30 can be provided.

The operation of the control module 130 including the model generating unit 100 , the collecting unit 200 , the selecting unit 300 , and the information providing unit 400 will be described in more detail below.

Next, a configuration of a detection model (DM) that is a deep learning model (DLM) according to an embodiment of the present invention will be described. 6 is a block diagram illustrating the configuration of a detection model (DM) that is a deep learning model (DLM) according to an embodiment of the present invention. 7 to 9 are diagrams for explaining the detailed configuration of a detection model (DM) that is a deep learning model (DLM) according to an embodiment of the present invention.

Referring to FIG. 6 , when analysis data including an image indicating the location of a link and a text in which a link is applied and a text to which a link is not applied is input, the detection model DM is disclosed based on the analysis data This is to calculate a probability indicating whether product sales information is included in a chat room or channel as an output value. This detection model (DM) includes an image detection network (IDN), a text detection network (TDN), a combiner (ADD), and a merge detection network (MDN).

The image detection network (IDN) is a product in a public chat room or channel corresponding to the input image through a plurality of operations in which a plurality of inter-layer weights are applied to the input image when an image with the area of the link (LK) is input. A first detection vector DV1 indicating the probability of whether or not sales information is included is calculated and output.

The text detection network (TDN) is a text that corresponds to the input text through a plurality of operations in which a plurality of inter-layer weights are applied to the input text when a text in which a text to which a link is applied and a text to which a link is not applied are separated is input. A second detection vector DV2 representing the probability of whether product sales information is included in the public chat room or channel is calculated and output.

The combiner (ADD) generates a merge detection vector (MDV) by merging the first detection vector (DV1) and the second detection vector (DV2), and inputs the generated merge detection vector (MDV) to the merge detection network (MDN). do.

When a merge detection vector (MDV) including a first detection vector (DV1) and a second detection vector (DV2) is input to the merge detection network (MDN), a weight between a plurality of layers is applied to the merge detection vector (MDV). An output value (T/F) indicating the probability of whether product sales information is included in a public chat room or channel corresponding to the input analysis data through a plurality of calculations is calculated and output.

Referring to FIG. 7 , an image detection network (IDN) according to an embodiment of the present invention may typically exemplify a Convolutional Neural Network (CNN). The image detection network (IDN) is sequentially an input layer (IL), a convolution layer (CL), a pooling layer (PL), a fully-connected layer (FL), and an output layer ( output layer: OL). Here, each of the convolutional layer CL, the pooling layer PL, and the fully connected layer FL may be two or more. The convolution layer CL and the pooling layer PL include at least one feature map (FM). The feature map FM is derived as a result of receiving a value applied with a weight and a threshold to the operation result of the previous layer, and performing an operation on the input value. These weights are applied through a filter or kernel W, which is a weight matrix of a predetermined size. In the embodiment of the present invention, the first filter W1 is used for the convolution operation of the convolutional layer CL, and the second filter W2 is used for the pooling operation of the pooling layer PL.

When an image including a link LK is input to the input layer IL, the convolution layer CL performs a convolution operation and an activation function on the image of the input layer IL using the first filter W1. At least one first feature map FM1 is derived by performing an operation by . Next, the pooling layer VPL performs a pooling or sub-sampling operation using the second filter W2 on at least one first feature map FM1 of the convolution layer VCL to obtain at least one A second feature map FM2 is derived.

The final connection layer FL includes a plurality of operation nodes E1 to En. The plurality of operation nodes E1 to En of the final connection layer FL are calculated by an activation function for at least one second feature map FM2 of the pooling layer PL, and a plurality of operation node values E[e1 , e2, e3, ..., en].

The output layer OL includes two output nodes Ox1 and Ox2. Each of the plurality of operation nodes E1 to En of the fully connected layer VFL is connected to the output nodes Ox1 and Ox2 of the output layer OL through a channel having a weight (W). In other words, the plurality of operation node values e1, e2, e3, ..., en of the plurality of operation nodes E1 to En are weighted and input to the output nodes Ox1 and Ox2, respectively. Accordingly, the output nodes Ox1, Ox2 of the output layer OL are calculated by the activation function for the plurality of operation node values e1, e2, e3, ..., en to which the weight is applied to the input layer IL ), calculates and outputs a first detection vector DV1 indicating the probability that the product sales information will be included and the probability that the product sales information will not be included in the open chat room or channel corresponding to the input image. Each of the two output nodes Ox1 and Ox2 of the output layer OL corresponds to a probability that product sales information is included and a probability that product sales information is not included. That is, the first output node Ox1 corresponds to the probability that the product sales information is included, and the second output node Ox2 corresponds to the probability that the product sales information is not included. Accordingly, the output value of the first output node Ox1 represents the probability that product sales information is included in the public chat room or channel corresponding to the input image, and the output value of the second output node Ox2 is the public corresponding to the input image. It indicates the probability that product sales information is not included in the chat room or channel.

Activation functions used in the above-described convolutional layer (CL), final connection layer (FL) and output layer (OL) are Sigmoid, Hyperbolic tangent (tanh), Exponential Linear Unit (ELU), and ReLU. (Rectified Linear Unit), Leakly ReLU, Maxout, Minout, Softmax, etc. may be exemplified. Any one of these activation functions may be selected and applied to the convolutional layer CL, the final connection layer FL, and the output layer OL.

Referring to FIG. 8 , the text detection network (TDN) may be a multilayer perceptron (MLP) among deep learning models. A text detection network (TDN) includes a plurality of layers (IL, HL, OL). The plurality of layers includes an input layer (IL), a plurality of hidden layers (HL1 to HLk), and an output layer (OL). In addition, each of the plurality of layers IL, HL, and OL includes a plurality of nodes. For example, as illustrated, the input layer IL may include n input nodes i1 to in, and the output layer OL may include two output nodes o1 and o2. Also, the number of hidden layers HL may be k (HL1, HL2, ..., HLk). Among the hidden layers HL, the first hidden layer HL1 includes a number of hidden nodes h11 to h1a, the second hidden layer HL2 includes b number of hidden nodes h21 to h2b, and the kth hidden layer HLk ) may include z hidden nodes (hk1 to hkz).

All of the plurality of nodes of the plurality of layers have operations. These operations are performed through the activation function. In particular, a plurality of nodes of different layers are connected by a channel (indicated by a dotted line) having a weight (w: weight). In other words, the calculation result of one node is weighted and becomes the input of the next layer node. That is, any one node of any one layer of the text detection network (TDN) receives a value obtained by applying a weight to the input from the node of the previous layer, sums them up, takes an activation function, and uses this result as an input of the next layer. transmit

Accordingly, when text in which text to which links are applied and text to which links are not applied is input, a plurality of operations in which weights between a plurality of layers (IL, HL, OL) are applied to the input text are performed. A second detection vector DV2 indicating a probability that product sales information is included and a probability that product sales information is not included in an open chat room or channel corresponding to the text is calculated and output.

More specifically, when text is input to the input layer IL of the text detection network TDN, the plurality of hidden layers HL1, HL2, ..., HLk from the input layer IL for the input pseudo data or reservation data. ) and the output layer OL by performing a plurality of calculations to which weights are applied to calculate the second detection vector DV2. Each of the two output nodes Oy1 and Oy2 of the output layer OL corresponds to a probability that product sales information is included and a probability that product sales information is not included. That is, the first output node Oy1 corresponds to the probability that the product sales information is included, and the second output node Oy2 corresponds to the probability that the product sales information is not included. Accordingly, the output value of the first output node Oy1 represents the probability that product sales information is included in the public chat room or channel corresponding to the input text, and the output value of the second output node Oy2 is the public corresponding to the input text. It indicates the probability that product sales information is not included in the chat room or channel.

Referring to FIG. 9 , the merge detection network (MDN) includes a fully connected layer (FCL) and a detection layer (DL). The fully connected layer FCL may include a plurality of operation nodes f1 to f4 , and the detection layer DL may include two detection nodes D1 and D2 . Likewise, all of the plurality of nodes of the plurality of layers of the merge detection network (MDN) have operations. That is, in the entire detection network (DN), any one node of any one layer receives a value obtained by applying a weight to the input from the node of the previous layer, sums them up, takes an activation function, and uses this result as the input of the next layer. transmit Since the description of the operation to which these weights are applied is the same as the description of the above-described image detection network (IDN) and text detection network (TDN), a more detailed description will be omitted.

When a merge detection vector (MDV) including the first detection vector (DV1) and the second detection vector (DV2) is input to the fully connected layer (FCL) of the merge detection network (MDN), the input merge detection vector (MDV) ), the fully connected layer FCL and the detection layer DL perform a plurality of operations in which weights are sequentially applied to calculate a detection value. Here, each of the two detection nodes D1 and D2 of the detection layer DL corresponds to a probability that product sales information is included and a probability that product sales information is not included in an open chat room or channel corresponding to the input analysis data. That is, the first detection node D1 corresponds to the probability that the product sales information is included, and the second detection node D2 corresponds to the probability that the product sales information is not included. Accordingly, the output value of the first detection node D1 indicates the probability that the product sales information is included, and the output value of the second detection node D2 indicates the probability that the product sales information is not included. For example, each of the detection nodes D1 and D2 applies a weight w=[w1, w2, w3, w4] to each of the plurality of operation node values f1 to f4, and then takes the activation function F to the result and calculates an output value do. For example, if the output values of the first and second detection nodes D1 and D2 are 0.753 and 0.247, respectively, the probability of including product sales information in a public chat room or channel corresponding to the analysis data including the input image and text is 75 %, indicating that the probability that product sales information is not included in the public chat room or channel corresponding to the analysis data including the input image and text is 25%. The detection values of the first and second detection nodes D1 and D2 of the detection layer DL become output values of the entire detection network DN. According to the output value of the detection network DN, the collection unit 200 may determine whether product sales information is included in the corresponding public chat room or channel. That is, as described above, if the output values of the first and second detection nodes D1 and D2 are 0.753 and 0.247, respectively, the probability that the product sales information is included is 75%, and the probability that the product sales information is not included is 25 %, so it is judged that product sales information is included in the relevant public chat room or channel.

Next, the user device 30 according to an embodiment of the present invention will be described. 10 is a block diagram for explaining the configuration of a user device according to an embodiment of the present invention. Referring to FIG. 10 , the user device 30 according to the embodiment of the present invention includes a communication unit 31 , an input unit 32 , a display unit 33 , a storage unit 34 , and a control unit 35 .

The communication unit 31 is for communication with other devices including the sales information server 10, and performs communication through various communication connection methods. The communication unit 31 may select and communicate with any one of communication functions of various communication connection methods. The communication unit 31 may be implemented as one module or a plurality of modules. For example, the communication unit 31 is a wireless local area network (WLAN) method using a broadband mobile communication method that accesses a network through a base station according to standards such as WCDMA, LTE, LTE-A, 5G, and wireless fidelity (Wi-Fi) method. Accordingly, a wireless short-range communication method for performing communication by accessing a network through an access point (AP) may be used. The communication unit 31 may include an RF transmitter for up-converting and amplifying a frequency of a transmitted signal, and an RF receiver for low-noise amplifying and down-converting a received signal. The data transmitted from the controller 35, for example, a search word for searching for sale product information, or an arrangement condition, may be converted into a wireless signal and transmitted through a wireless channel. In addition, the communication unit 31 may receive data, for example, a wireless signal including a list of sales product information through a wireless channel, and transmit it to the control unit 35 .

The input unit 32 receives a user's key manipulation for controlling the user device 30 , generates an input signal, and transmits it to the control unit 35 . The input unit 32 may include any one of a power key, a number key, and a direction key for power on/off, and may be formed as a predetermined function key on one surface of the user device 30 . When the display unit 33 is formed of a touch screen, the functions of various keys of the input unit 32 may be performed on the display unit 33, and when all functions can be performed only with the touch screen, the input unit 32 may be omitted. may be

The display unit 33 visually provides a menu of the user device 30 , input data, function setting information, and other various information to the user. The display unit 33 performs a function of outputting various screens such as a booting screen, a standby screen, and a menu screen of the user device 30 . The display unit 33 may be formed of a liquid crystal display (LCD), an organic light emitting diode (OLED), an active matrix organic light emitting diode (AMOLED), or the like. Meanwhile, the display unit 33 may be implemented as a touch screen. In this case, the display unit 33 may include a touch sensor, and the control unit 35 may sense a user's touch input through the touch sensor. The touch sensor may be composed of a touch sensing sensor such as a capacitive overlay, a pressure type, a resistive overlay, or an infrared beam, or may be composed of a pressure sensor. . In addition to the above sensors, all kinds of sensor devices capable of sensing contact or pressure of an object may be used as the touch sensor of the present invention. The touch sensor detects a user's touch input, generates a detection signal, and transmits it to the controller 35 . The detection signal may include coordinate data input by the user. When the user inputs the touch position movement motion, the touch sensor may generate a detection signal including coordinate data of the touch position movement path and transmit it to the controller 35 .

The storage unit 34 serves to store programs and data necessary for the operation of the user device 30 , and may be divided into a program area and a data area. The program area may store a program for controlling the overall operation of the user device 30 , an operating system (OS) for booting the user device 30 , an application program, and the like. The data area is an area in which data generated according to the use of the user device 30 is stored. Also, the storage unit 34 may store various types of data generated according to the operation of the user device 30 .

The controller 35 may control the overall operation of the user device 30 and the signal flow between internal blocks of the user device 30 , and may perform a data processing function of processing data. The controller 35 may be a central processing unit (CPU), an application processor (AP), a graphic processing unit (GPU), or the like.

When a search word for searching for a product to be purchased and an alignment condition for arranging product sales information for the product are input by the user through the input unit 32, the control unit 35 transmits the inputted search word, sort condition, etc. to the communication unit ( 31) through the sales information server 10 . Accordingly, the control unit 35 may receive a list of product sales information corresponding to the search word and the sort condition from the sales information server 10 , and display the list of the received product sales information through the display unit 33 .

Next, a method for collecting and providing product sales information posted on a mobile messenger according to an embodiment of the present invention will be described. 11 is a flowchart illustrating a method for collecting and providing product sales information posted on a mobile messenger according to an embodiment of the present invention.

Referring to FIG. 11 , the collection unit 200 connects to the messenger server 20 through the communication module 11 in step S110 and uses a detection model (DM) that has completed learning to include a plurality of products including sales information. Detect public chat rooms and channels of A method of detecting a plurality of open chat rooms and channels including product sales information will be described in more detail below.

Next, the collection unit 200 collects a plurality of product sales information from the plurality of public chat rooms and channels detected in step S120 . Here, the product sales information includes a product for sale, a seller, a sales volume, a sales price, a discount rate, a sales period, a recent sales date, a sales frequency, a link to a sales page, a link to a payment page, and the like.

Next, in step S130 , the selection unit 300 selects a plurality of product sales information whose reliability is greater than or equal to a predetermined level from the plurality of product sales information. A specific procedure for selecting a plurality of product sales information having a reliability level equal to or higher than a predetermined level will be described in more detail below.

Next, the selection unit 300 adds or updates the plurality of product sales information selected in step S140 to the storage module 12 and stores it. That is, the selection unit 300 compares the selected product sales information with the product sales information stored in the storage module 12 to newly add or update the product sales information. For example, when the existing product sales information exists in the storage module 12 , the selection unit 300 updates the sales price with the latest sales price when the sales price is changed. In addition, the selection unit 300 updates the sales volume by the amount accumulated in the existing sales volume by the amount newly sold. In addition, the selection unit 300 updates the pre-stored most recent sale date as the most recent sale date. As another example, when there is no existing product sales information in the storage module 12 , the selection unit 300 newly adds the selected product sales information to the storage module 12 .

Then, in step S150 , the information providing unit 400 may provide the sales information of a plurality of sellers stored according to the request of the user device 30 as a recommendation list. For example, the information providing unit 400 provides a plurality of product sales information according to a request of the user device 30 , that is, an arrangement condition, at least among a plurality of parameters including a selling price, a discount rate, a sales volume, a recent sales date, and a sales frequency. A list of product sales information may be configured and provided by sorting in ascending or descending order for one. In particular, in the case of sorting using a plurality of parameters, a list of product sales information can be configured by setting a priority for the parameter according to the request of the user device 30 and arranging the product sales information according to the set priority. have. For example, firstly, the product sales information is sorted in the order of the lowest selling price, if the sales price is the same, secondly, the product sales information is sorted in the order of the highest sales, and if the sales are the same, thirdly, the product sales A list of recommendations can be organized by arranging the information in the order of the most recent sales date.

Next, a method of detecting a plurality of open chat rooms and channels including product sales information using the detection model of step S110 described above will be described in more detail. First, deep learning of the detection model should be performed using the detection model. Accordingly, a deep learning method of the detection model DM will be described first. 12 is a flowchart for explaining a deep learning method for a detection model that is a deep learning model according to an embodiment of the present invention.

Prior to the description of the flowchart of FIG. 12 , in an embodiment of the present invention, the model generator 100 performs learning through a loss function as shown in Equation 1 below.

In Equation 1, CE means loss. λ is a hyperparameter. λ may have a range of a minimum value and a maximum value. For example, it may have a range of 0.5 < λ ≤ 2. i is an index corresponding to the training data. Li is a label corresponding to the i-th training data. In addition, DV1i is a first detection vector that is an output of the image detection network (IDN) corresponding to the i-th training data, and DV2i is a second detection vector that is an output of the text detection network (TDN) corresponding to the i-th training data. And Di denotes an output value of the merge detection network (MDN) corresponding to the i-th training data.

Referring to FIG. 12 , the model generator 100 prepares training data in step S210 . Here, the learning data is, for example, as shown in FIGS. 2 and 3, an image in which the area of the link (LK) is divided and displayed in the image of the public chat room or channel, and the text to which the link is applied in the text of the public chat room or channel, and Non-linked text contains delimited text.

Then, the model generation unit 100 sets the value of the hyperparameter λ in step S220. The model generator 100 increases the value of the hyperparameter λ by a predetermined numerical value each time the learning procedure is repeated using other learning data starting from the initial value (eg, 0.5) at the beginning of the learning, and sets the maximum value ( For example, when 2) is reached, the maximum value is maintained even when the learning procedure is repeated. This is to prevent overfitting.

Next, the model generation unit 100 sets a label on the training data in step S230. The model generator 100 sets a label for the training data by applying a one-hot encoding vector. The model generating unit 100, for example, in the case of the experimental group that is an image and text of a public chat room or channel including product sales information among the learning data, the label is set to [1, 0], and product sales information is not included. In the case of a control group that is an image and text of a non-public chat room or channel, the label is set to [0, 1].

Then, when the model generator 100 inputs the training data to the detection model DM, the detection model DM performs a plurality of calculations in which weights between a plurality of layers are applied to the training data input in step S240. The image detection network (IDN) and the text detection network (TDN) respectively calculate the first detection vector (DV1) and the second detection vector (DV2), and the combiner (ADD) the first detection vector (DV1) and the second detection Merging vectors (DV2), merge detection network (MDN) includes product sales information in public chat rooms or channels from merge detection vector (MDV) in which first detection vector (DV1) and second detection vector (DV2) are merged It calculates an output value indicating the probability of whether or not it is.

Then, in step S250, the model generator 100 uses a loss function as in Equation 1 to minimize the loss, which is the difference between the output value and the label, through a back-propagation algorithm to minimize the weight ( Perform optimization to correct w). Steps S210 to S250 described above are repeatedly performed using a plurality of different pieces of learning data. This iteration may be performed until the accuracy is calculated through the evaluation index, and a desired accuracy is reached.

Next, as described above, when the learning of the detection model (DM) is completed, as in the above-described step S110, the collection unit 200 uses the detection model (DM) to disclose a plurality of products including product sales information. Chat rooms and channels can be detected. These methods will be described. 13 is a flowchart for explaining a method of detecting a plurality of public chat rooms and channels including product sales information through a detection model according to an embodiment of the present invention.

Referring to FIG. 13 , the collection unit 200 connects to the messenger server 20 through the communication module 11 in step S310 and detects a link (LK), an image, and a text in an open open chat room and channel. Then, the collection unit 200 displays the area of the link (LK) on the image in step S320, distinguishes the text to which the link is applied from the text to which the link is not applied in the text, and links ( LK) area is divided and displayed, and analysis data including text in which a link is applied and a text to which a link is not applied are separated from the text of a public chat room or channel. Then, the collection unit 200 inputs the analysis data to the detection model DM in step S330.

Then, the detection model DM performs a plurality of calculations in which weights between a plurality of layers are applied to the analysis data input in step S340 so that each of the image detection network (IDN) and the text detection network (TDN) is the first detection vector (DV1) and the second detection vector (DV2) are calculated, the combiner (ADD) merges the first detection vector (DV1) and the second detection vector (DV2), and the merge detection network (MDN) merges the first detection vector From the merged detection vector (MDV) in which (DV1) and the second detection vector (DV2) are merged, an output value indicating the probability of whether product sales information is included in the corresponding open chat room or channel is calculated.

Then, the collection unit 200 determines whether product sales information is included in the corresponding open chat room or channel according to the probability indicated by the output value calculated in step S350 . For example, if the output values are 0.753 and 0.247, the probability that product sales information will be included in the public chat room or channel corresponding to the input analysis data is 75%, and product sales information is not included in the public chat room or channel corresponding to the input analysis data. This indicates that there is a 25% chance that it will not. Therefore, the collection unit 200 determines that the product sales information is included in the public chat room or channel because there is a higher probability that the product sales information is included.

Next, a method of selecting product sales information having a reliability level greater than or equal to a predetermined level from the plurality of product sales information in step S130 will be described in more detail. 14 is a flowchart for explaining a method of selecting a plurality of product sales information having a reliability level equal to or higher than a predetermined level from a plurality of product sales information according to an embodiment of the present invention. 15 is a diagram for explaining a method of embedding and clustering a plurality of product sales information in a vector space according to an embodiment of the present invention. FIG. 16 is a diagram for explaining a method of selecting product sales information having a reliability higher than or equal to a predetermined level from a plurality of product sales information according to an embodiment of the present invention.

Referring to FIG. 14 , the selection unit 300 generates a plurality of product sales information vectors by embedding a plurality of product sales information in a predetermined vector space VS in step S410 . The product sales information vector is a two-dimensional (rank) rank tensor representing the selling price, discount rate, sales volume, and recent sales date of the product to be sold. After generating the product sales information vector, the selection unit 300 randomly selects a center vector of a predetermined number of clusters from among a plurality of product sales information vectors in the vector space VS in step S420 .

When the center vector of each cluster is selected, the sorting unit 300 selects a plurality of product sales information vectors in the vector space VS as shown in FIG. 15 in step S430 according to a distance from the center vector of a predetermined number of clusters. A plurality of clusters CL1 , CL2 , and CL3 are generated by clustering. That is, in step S430 , the selection unit 300 calculates a distance to each of the center vectors of a predetermined number of clusters of a plurality of product sales information vectors in the vector space VS as shown in FIG. 15 , and sells a plurality of products. Each information vector is included in the cluster with the smallest distance from the center vector of the cluster.

Next, the selector 300 selects the center vector of the cluster again for each of the plurality of clusters CL1 , CL2 , and CL3 generated in step S440 . For example, in step S440 , the selection unit 300 may select a product sales information vector having a median value among a plurality of product sales information vectors of each cluster as a central vector. Then, in step S450 , the selection unit 300 checks whether all of the product sales information vectors are included in the cluster having the minimum distance from the center vector of the previously selected cluster again ( S440 ).

As a result of checking in step S450, if all of the product sales information vectors are not included in the cluster having the minimum distance from the center vector of the previously calculated cluster again (S440), the sorting unit 300 performs steps S430 to S450 described above. Repeat. In other words, the selection unit 300 repeats steps S430 to S450 until the following Equation 2 becomes the minimum.

In Equation 2, k is the index of the cluster, and Ck is the center vector of the k-th cluster. n denotes an index of the product sales information vector, and Sn denotes the nth product sales information vector. fnk is a flag variable having 1 if the nth product sales information vector belongs to the kth cluster, and 0 otherwise.

On the other hand, as a result of checking in step S450, if all the product sales information vectors are included in the cluster having the minimum distance from the center vector of the previously selected cluster again (S440), the selection unit 300 selects the current cluster in step S460. Determine the optimized cluster.

Then, the selector 300 derives a convex hull from each of the plurality of clusters optimized in step S470 . For example, one optimized cluster is shown in FIG. 16 . As shown, the selector 300 may configure a convex hull by selecting a plurality of convex from among a plurality of product sales information vectors included in the cluster. Next, the selection unit 300 sets the center of gravity of the polygon constituting the convex hull as the center of the cluster in step S480 . For example, the selection unit 300 may set the center of gravity of the hexagon constituting the convex hull as the center of the cluster, as shown in FIG. 16 . Although the product sales information vector may not exist at the center of the cluster, the center of the cluster may be expressed as the same vector value as the product sales information vector.

Next, in step S490 , the selection unit 300 calculates a loss representing a difference between the center of a cluster and a plurality of product sales information vectors in each of the plurality of clusters, and calculates a critical loss (T) from the average and standard deviation of the losses. . More specifically, the selection unit 300 calculates a loss E representing a difference between the center Mk of the cluster and the plurality of product sales information vectors Sn according to Equation 3 below.

In Equation 3, E represents the loss, k is the index of the cluster, and Mk represents the center of the k-th cluster. n is the index of the product sales information vector, and Sn means the nth product sales information vector.

Then, the selection unit 300 calculates the critical loss (T) from the average and standard deviation of the loss according to the following Equation (4).

는 표준편차 에 대한 가중치이며,

의 범위를 가지며, 모집단의 크기에 비례하여 미리 설정되는 하이퍼파라미터이다. 산출된 임계손실(T)은 각 클러스터 별로 저장된다. In Equation 4, T denotes a critical loss. In addition, each of w and σ represents the average and standard deviation of the loss E representing the difference between the center Mk of the cluster and the plurality of product sales information vectors Sn calculated according to Equation 3 above. In addition,

is the weight for the standard deviation,

It is a hyperparameter that has a range of , and is preset in proportion to the size of the population. The calculated critical loss (T) is stored for each cluster.

As described above, after calculating the critical loss (T), the selection unit 300 extracts a product sales information vector having a loss (E) equal to or less than the critical loss (T) from each of the plurality of clusters in step S500 . For example, as shown in FIG. 16 , a product sales information vector having a loss E less than or equal to the critical loss T means a product sales information vector within a radius defined by the critical loss T from the center of gravity.

Next, the selection unit 300 selects a seller corresponding to the product sales information vector having a loss E equal to or less than the critical loss T extracted from each of the plurality of clusters as a seller having a reliability level greater than or equal to a predetermined level in step S510 . For example, as shown in FIG. 16, a product sales information vector having a loss E less than or equal to the critical loss T corresponds to a product sales information vector within a radius defined by the critical loss T from the center of the cluster. This means selecting sellers with a reliability level higher than or equal to a certain level.

Meanwhile, the method according to the embodiment of the present invention described above may be implemented in the form of a program readable by various computer means and recorded in a computer readable recording medium. Here, the recording medium may include a program command, a data file, a data structure, etc. alone or in combination. The program instructions recorded on the recording medium may be specially designed and configured for the present invention, or may be known and available to those skilled in the art of computer software. For example, the recording medium includes magnetic media such as hard disks, floppy disks and magnetic tapes, optical recording media such as CD-ROMs and DVDs, and magneto-optical media such as floppy disks ( magneto-optical media) and hardware devices specially configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions may include not only machine language wires such as those generated by a compiler, but also high-level language wires that can be executed by a computer using an interpreter or the like. Such hardware devices may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

Although the present invention has been described above using several preferred embodiments, these examples are illustrative and not restrictive. As such, those of ordinary skill in the art to which the present invention pertains will understand that various changes and modifications can be made in accordance with the doctrine of equivalents without departing from the spirit of the present invention and the scope of rights set forth in the appended claims.

10: Sales information server 11: Communication module
12: storage module 13: control module
20: messenger server 30: user device
31: communication unit 32: input unit
33: display unit 34: storage unit
35: control unit 100: model generation unit
200: collection unit 300: sorting unit
300: information provider

Claims (4)

In the device for collecting and providing product sales information posted on a mobile messenger,
a collection unit for collecting a plurality of product sales information from at least one of open chat rooms and channels opened by accessing the messenger server;
a selection unit configured to select a plurality of product sales information whose reliability is higher than or equal to a predetermined level from the plurality of product sales information; and
a recommendation unit for arranging the selected plurality of product sales information according to a request of the user device to configure a recommendation list, and providing the configured recommendation list to the user device;
characterized in that it comprises
A device for collecting and providing product sales information. According to claim 1,
the collection unit
Detects links, images and texts in the public chat room and the channel,
Generates analysis data including an image in which the area of the link is divided and displayed in the image, and text in which the text to which the link is applied and the text to which the link is not applied are separated from the text;
Input the analysis data to the detection model,
The detection model calculates a first detection vector and a second detection vector by performing a plurality of calculations in which a weight between a plurality of layers is applied to the analysis data, and by merging the first detection vector and the second detection vector When a merge detection vector is derived and an output value representing the probability of whether product sales information is included in a public chat room or channel is calculated from the merge detection vector,
characterized in that it is determined whether product sales information is included in the corresponding open chat room or channel according to the probability indicated by the calculated output value
A device for collecting and providing product sales information. According to claim 1,
The selection unit
A plurality of product sales information is embedded in a predetermined vector space to generate a plurality of product sales information vectors,
A plurality of product sales information vectors are clustered into a plurality of clusters, and a convex hull is derived from each of the clustered plurality of clusters,
In each of the plurality of clusters, the center of gravity of the polygon constituting the convex hull is set as the center of the cluster,
for each cluster
formula

Calculate the loss representing the difference between the center of the cluster and the plurality of product sales information vectors through
From the mean and standard deviation of the loss representing the difference between the center of the cluster and the plurality of product sales information vectors for each of the plurality of clusters
formula

Calculate the critical loss according to
extracting a product sales information vector having a loss equal to or less than a threshold loss calculated from each of the plurality of clusters, and selecting a seller corresponding to the extracted product sales information vector as a seller having a reliability level higher than or equal to a predetermined level;
where n is an index,
E represents the loss,
where k is the index of the cluster,
Mk represents the center of the k-th cluster,
Where n is the index of the product sales information vector,
wherein Sn is the nth product sales information vector,
where T is the critical loss,
where w is the average of the loss representing the difference between the center of the cluster and the plurality of product sales information vectors,
σ is the standard deviation of the loss representing the difference between the center of the cluster and the plurality of product sales information vectors,
remind

is a weight for the standard deviation, characterized in that
A device for collecting and providing product sales information. A method for collecting and providing product sales information posted on a mobile messenger,
collecting, by a collecting unit, a plurality of product sales information from at least one of open chat rooms and channels opened by accessing a messenger server;
selecting, by a selection unit, a plurality of product sales information having a reliability level higher than or equal to a predetermined level from the plurality of product sales information; and
arranging the selected plurality of product sales information according to the request of the user device by the information providing unit to configure a list of product sales information, and providing the configured list of product sales information to the user device;
characterized in that it comprises
A method for collecting and providing product sales information.

Images