KR101926403B1 - Promotion Performance Prediction and Recommendation Apparatus in Online Shopping Mall Using Artificial Intelligence - Google Patents

Abstract

The present invention relates to a promotion performance prediction and recommendation apparatus in an online shopping mall using artificial intelligence and a method thereof. According to an embodiment of the present invention, the promotion of the online shopping mall becomes elaborate and precise by the learning of the artificial intelligence, thereby raising a promotion effect. The promotion performance prediction and recommendation apparatus includes a memory module for storing a program code and a processing module for predicting the performance of a specific promotion and processing a promotion attribute.

Description

{Promotion Performance Prediction and Recommendation Apparatus in Online Shopping Mall Using Artificial Intelligence}

The present invention relates to an apparatus and a method for predicting and recommending promotion performance in an online shopping mall using artificial intelligence.

Promotions in online shopping malls are in the form of product discounts, sweepstakes events, discount coupon payments, and mileage payments. Because these promotions always lead to a decline in net profit from an online shopping mall, the type of promotions, the duration of promotions, and the degree of promotion must be precisely designed to match the persona. In the past, however, the design of promotions has been determined empirically by the operators of online shopping malls and promotional agencies.

There are various forms of online shopping in Korea such as open market, social commerce, general shopping mall, large mart shopping mall, and card point mall. Most sellers are selling in multiple online shopping malls, It is difficult to analyze the sales performance based on the collected data.

The components of a promotion are related to the date and condition of the promotion, which is causal to the promotion result. However, various items such as season, month, day, time, It is very difficult for a person to calculate the contribution of the component based on the result.

In particular, in the case of promotional depreciation, which will serve as a benefit to consumers, the greater the depreciation rate, the greater the purchase benefits. As a result, sales volume will increase and sales will increase. However, considering the cost of the product, it is important to balance the sales volume increase and the depreciation rate at the appropriate point, and it is very difficult for a person to calculate whether or not the expected sales profit will be achieved .

A Web advertisement recommendation method and system based on a click pattern using a collaborative filtering system having a neural network, Korean Patent Registration No. 10-0792700, Naver Corporation

Accordingly, it is an object of the present invention to provide an online shopping mall using artificial intelligence that predicts the performance of a promotion using artificial intelligence and recommends a characteristic of the promotion, in order to solve the low precision of the promotion of the online shopping mall, And to provide a recommendation apparatus and method for predicting and recommending the promotional performance of the promotional products.

Hereinafter, specific means for achieving the object of the present invention will be described.

An object of the present invention is to provide a memory module that stores program codes of an artificial neural network learned based on product attribute data, promotion attribute data, other performance data, and promotions; And a processing module for processing the program code of the artificial neural network to predict the performance of a specific promotion; Wherein the program code comprises: an input step of receiving product attribute data (100), promotion attribute data (200), and performance data (300); A performance prediction step of predicting performance of the specific promotion on the basis of the product property data, the promotion property data, and the performance data to generate the promotion performance data; And an output step of outputting the predicted promo performance data. The present invention can be achieved by providing a promotional performance prediction and recommendation apparatus in an online shopping mall using artificial intelligence configured to be performed on a computer.

A memory module for storing program codes of the promotional performance prediction artificial neural network module and the promotion recommendation module, which have been previously learned based on the promotion attribute data and the performance data; And a processing module processing the program code of the promotion performance prediction artificial neural network module and the promotion recommendation module to predict a performance of a specific promotion and to recommend a promotion attribute, wherein the program code of the promotion recommendation module includes: A performance prediction predicting step of receiving promotional performance prediction data, which is information on predicted promotion performance for a specific promotion attribute in the performance prediction artificial neural network module; A status information receiving step of receiving status information (s _t ) for episode t; Generating a compensation probability information in a value network based on the state information and the promotion performance prediction data and outputting an attribute probability output for outputting a recommended promotion property probability as a plurality of actions in a policy network based on the compensation possibility information step; A recommended promotional attribute data output step of selecting and outputting recommended promotional attribute data (a _t ) for the episode t through an optimum promotional search based on the plurality of recommended promotional attribute probabilities; Receiving state information and compensation information (rt _{+ 1} ) for state information (s _{t + 1} ) and a _t for episode t + 1 after the promotion according to the recommended promotional attribute data is applied; And updating the policy network and the value network based on the received status information (s _{t + 1} ) and the compensation information (r _{t + 1} ); Wherein the status information is configured to be executed on a computer, the status information including at least one of performance data (such as a quantity, a sales amount, a page view, a conversion rate, an average sales price, Sales promotion information, sales profit) and environmental data (date information, day of the week information, weather information, promotional product information, shopping mall information, or other product information of the shopping mall), and the recommended promotion attribute data includes promotional category, Wherein the compensation information refers to a color of an exposure account or a promotion image and the compensation information means a rise or a fall of the performance data for a specific period of time. Can be achieved.

The optimal promotional search in the step of outputting the recommended promotional attribute data may be performed using a Monte Carlo tree search. In the Monte Carlo tree search, each node of the tree may represent a state, each edge may be a specific action And a leaf node is expanded each time a transition is made to a new state by taking a new action with the current state as a root node, and the recommended promotional attribute data output step A selection step of selecting an action having the highest value among the selectable actions from the current state until the leaf node comes out; An extension step of adding a new leaf node according to a probability of the policy network learned by the learning of the guidance when the simulation proceeds to the leaf node; An evaluation step of evaluating the value of the new leaf node by calculating the compensation possibility information determined using the value network from the new leaf node and the compensation obtained from the leaf node until the end of the promotion episode using the policy network; And an update step of re-evaluating the value of visited nodes during simulation and reflecting the value of the new leaf node, and updating the visit frequency, wherein the recommendation promotion attribute data (a _t ) is selected based on the visit frequency .

In addition, the policy network may be configured to perform map learning based on the promotion attribute data and the performance data including a random vector.

Another object of the present invention is to provide a promotional performance prediction step of receiving promotional performance prediction data, which is information on a promotional performance predicted for a specific promotional attribute in a promotional performance artificial neural network module; A status information receiving step of the promotion recommendation module receiving status information (s _t ) for the episode t; The promotion recommendation module generates the compensation possibility information in the value network on the basis of the state information and the promotion performance prediction data, and then, based on the compensation possibility information, the recommendation promotion probability, which is a plurality of actions in the policy network, Outputting an attribute probability output step; A recommended promotional attribute data output step of selecting and outputting recommended promotional attribute data (a _t ) for the episode t through an optimum promotional search based on the plurality of recommended promotional attribute probabilities; After the promotion based on the recommended promotional attribute data is applied, the promotion recommendation module updates the state and compensation (rt _{+ 1} ) for state information (s _{t + 1} ) and a _t for episode t + Receiving information; And a policy network and a value network updating step of updating the policy network and the value network based on the status information (s _{t + 1} ) and the compensation information (r _{t + 1} ) received by the promotion recommendation module; Wherein the status information is configured to be executed on a computer, the status information including at least one of performance data (such as a quantity, a sales amount, a page view, a conversion rate, an average sales price, Sales promotion information, sales profit) and environmental data (date information, day of the week information, weather information, promotional product information, shopping mall information, or other product information of the shopping mall), and the recommended promotion attribute data includes promotional category, Wherein the compensation information refers to a color of an exposure account or a promotion image and the compensation information means a rise or a fall of the performance data for a specific period of time. Can be achieved.

As described above, the present invention has the following effects.

First, according to the embodiment of the present invention, the promotion of the online shopping mall is elaborated and refined by the learning of the artificial intelligence, so that the effect of the promotion is raised.

Second, according to an embodiment of the present invention, an update of the policy network that outputs the probability of a recommended promotional attribute according to the value network 111 can be effected for each episode. In the existing reinforcement learning, there is a problem that the update of the reinforcement learning model is carried out after all the episodes have ended, so that it has been difficult to apply to the recommendation recommendation model.

Thirdly, according to the embodiment of the present invention, since the recommendation of the promotion is short-term and long-term, there is difficulty in applying the DQN because the assumption of the MDP is not established, Promotional recommendation can also be reinforced.

Fourth, according to an embodiment of the present invention, the learning speed is improved because the policy network and the value network are connected to the promotion performance prediction artificial neural network device and learned based on the promotion performance data.

BRIEF DESCRIPTION OF THE DRAWINGS The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this application, illustrate preferred embodiments of the invention and, together with the description, And shall not be interpreted.
FIG. 1 is a diagram illustrating a promotion performance prediction and recommendation apparatus 1 in an online shopping mall using artificial intelligence according to an embodiment of the present invention,
FIG. 2 is a schematic diagram showing a learning process of a promotional performance prediction artificial neural network device according to an embodiment of the present invention;
3 is a schematic diagram illustrating a reasoning process of the promotional performance prediction artificial neural network apparatus according to an embodiment of the present invention;
4 is a schematic diagram showing a promotion recommendation apparatus 11 according to an embodiment of the present invention,
5 is a flowchart illustrating an example of the operation of the promotion recommendation apparatus 11 according to an embodiment of the present invention,
6 is a schematic diagram showing an apparatus 13 for revising a promotion image according to an embodiment of the present invention,
7 is a schematic diagram showing an example of a ConvNet encoder according to an embodiment of the present invention,
FIG. 8 is a flowchart illustrating a recommendation method according to an exemplary embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following detailed description of the operation principle of the preferred embodiment of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may unnecessarily obscure the subject matter of the present invention.

The same reference numerals are used for portions having similar functions and functions throughout the drawings. In the specification, when a specific portion is connected to another portion, it includes not only a direct connection but also a case where the other portion is indirectly connected with another element in between. In addition, the inclusion of a specific constituent element does not exclude other constituent elements unless specifically stated otherwise, but may include other constituent elements.

Promotion Performance Prediction and Recommendation Device and Method in Online Shopping Mall Using Artificial Intelligence

FIG. 1 shows a promotional performance prediction and recommendation apparatus 1 in an online shopping mall using artificial intelligence according to an embodiment of the present invention. As shown in FIG. 1, the promotion performance prediction and recommendation apparatus in the online shopping mall using artificial intelligence according to an embodiment of the present invention is configured to perform a promotion promotion based on product attribute data, promotion attribute data, performance data, A performance prediction artificial neural network device 10 and a promotion recommendation device 11.

FIG. 2 is a schematic diagram showing a learning process of a promotional performance prediction artificial neural network device according to an embodiment of the present invention. FIG. 3 is a diagram illustrating a promotional performance prediction according to an embodiment of the present invention. Fig. 2 is a schematic diagram showing an inference process of an artificial neural network device. Fig. As shown in FIGS. 2 and 3, the method for predicting promotional performance in an online shopping mall using artificial intelligence receives product attribute data 100, promotion attribute data 200, and performance data 300, It is possible to output the predicted promotional performance prediction data 400 by learning the apparatus 10 and inputting the product attribute data 100 and the promotion attribute data 200 to the previously learned promotional performance prediction ANN apparatus 10 have. In particular, during the learning of the promotion performance prediction artificial neural network device 10, the weight of the hidden layer may be updated based on the error between the promotion performance data and the output data. 2, each attribute constituting the product attribute data 100 and the promotion attribute data 200 is input to each node of the input layer such as x1, x2, and x3, and h1, h2, h3 The predicted promotion performance data 400 based on the cost function such as softmax is output to the output layer y1. The weight of the promotional performance prediction artificial neural network device 10 may be updated to back propagation based on the error (-Sigma (y _i log p _i )) between the predicted performance data and the actual performance data 300.

Alternatively, a separate promotional performance artificial neural network device 10 is configured for each product, and the promotional performance data 200 and the performance data 300 are received to learn the promo performance prediction artificial neural network device 10, The predicted promotional performance data 400 can be output by inputting the promotional attribute data 200 to the predicted promotional performance neural network device. Machine learning such as multi-layer perceptron, Naive-Bayesian classification, and random forest classification can be used in addition to the artificial neural network that can be used at this time.

The product attribute data 100 may include a product category, a product selling price, a product cost, and the like.

Promotion attribute data 200 may include promotional categories (discount promotions, gift promotions, etc.), promotion periods, promotional malls, exposure accounts, promotional images, and the like. Promotion image data among the promotion attribute data according to an embodiment of the present invention may mean a result value classified into a specific atmosphere category and specific color category through the CNN (Convolutional Neural Network) that has been learned.

The performance data 300 may include a quantity, a sales amount, a page view, a conversion rate, an average sales price, a sales profit, and the like. According to an embodiment of the present invention, in order to prevent a problem due to the multicollinearity of the performance data 300, PCA is applied to each data constituting the performance data 300 to form a diagnosis matrix It can offset the collinearity.

In connection with the recommendation of the promotional attribute, the promotional recommendation apparatus 11 may be configured to output the recommended promotional attribute data for outputting the highest promotional performance predictive data 400 using the promotional performance predictive NE.

Promotion recommendation of the promotion recommendation apparatus 11 according to an embodiment of the present invention can be performed using reinforcement learning. 4 is a schematic diagram showing a promotion recommendation apparatus 11 according to an embodiment of the present invention. 4, the promotion recommendation apparatus 11 according to an embodiment of the present invention includes a value network 111 as an artificial neural network that learns a value function for outputting a value in a specific state, The policy network 110 and the value network 111 according to an embodiment of the present invention may be connected to the promotion performance prediction ANN apparatus 10 Lt; / RTI > In addition, the policy network 110 and the value network 111 may be connected to the optimal promotion search module 12 to output the recommended promotion attribute data 500.

In view of the reinforcement learning, the objective of the promotion recommendation apparatus 11 according to the embodiment of the present invention is to improve the performance of the promotion, and the state is the current performance data (quantity of sales (Quantity), sales amount ), Inflow (Page View), conversion rate, average sales price, sales profit, etc.) and environmental data (date, day, weather, current promotional products, shopping mall, ), Action may mean promotional attribute data (promotional category (discount promotion, gift promotion, etc.), promotion period, promotion shopping mall, exposure account, promotion image color, etc.) ) May refer to an increase or decrease in performance data for a particular period of time.

The policy network 110 is an artificial neural network that determines the probability of specific attributes of a recommendation promotion in each state of the recommendation recommendation apparatus 11 and learns a policy function to output a recommended promotional attribute probability. The cost function of the policy network may be a function obtained by calculating the cross entropy by multiplying the policy function and the cost function of the value network, and then taking the policy gradient, for example, as shown in Equation 1 below. The policy network can be back propagated based on the product of the cross entropy and the time difference error, which is the cost function of the value network.

In the equation 1, π is a policy function, θ is a policy network parameter, π _θ (a _i │s _i ) is the probability that a particular action (promotion of a particular attribute) is present in the current episode, V is a value function, _I is the state information of the current episode i, S _{i + 1} is the state information of the next episode i + 1, r _{i + 1} is the compensation expected to be obtained in the next episode, V _w (s _i ) The compensation potential in the current episode, V _w (s _{i + 1} ), can be compensated for in the next episode, and γ can mean the depreciation rate. At this time, r _{i + 1} can be configured to be received at the promo performance prediction artificial neural network device 10. As a result, the policy network 110 according to an exemplary embodiment of the present invention outputs a promotional attribute that simulates a user's promotional history through a policy gradient.

The policy network 110 according to the embodiment of the present invention performs supervised learning based on the existing promotion attribute data and the performance data thereof before the reinforcement learning progresses to update the weight of the policy network, You can learn the basics. That is, the weight of the policy network can be learned and set based on the existing promotion attribute data and performance data. According to this, the policy network can be learned very quickly by the recording of the existing promotion.

In addition, according to an embodiment of the present invention, it is possible to configure the policy network 110 to perform map learning based on existing promotion attribute data including random vectors and performance data corresponding thereto. The random vector may use, for example, a Gaussian distribution. According to this, the policy network can output a challenging promotion policy with a random probability. If the map learning is performed based on the existing promotion attribute data and the performance data based on the existing promotion attribute data at the time of learning the policy network 110, the recommendation of the promotion is optimized in the existing policy. However, according to one embodiment of the present invention, when the random vector is included in the learning of the policy network, the policy network can learn a more effective promotion than the existing policy as the reinforcement learning progresses.

The value network 111 is an artificial neural network that derives the possibility of achieving a reward in each state that the promotion recommendation apparatus 11 can have and learns the value function. The value network 111 presents the direction of the direction in which the promotional recommendation apparatus 11 as an agent is to be updated. To this end, the input variable of the value network 111 is set to state information, which is information on the state of the promotion recommendation apparatus 11, and the output variable of the value network 111 is set to the state where the promotion recommendation apparatus 11 It is possible to set the possibility of compensation possibility information. The compensability information according to an exemplary embodiment of the present invention can be calculated by a Q-function such as the following equation.

In Equation (2), Q _π denotes the total compensability information expected in the future in the case of the state s and action a in the specific policy π, R can be a compensation for a specific period, and gamma can mean a depreciation rate. S _t is the state of time t, A _t is the action of time t, and E is the expected value. The Q value according to an embodiment of the present invention defines the update direction and size of the policy network 110. [

At this time, the cost function of the value network can be a mean square error (MSE) function for the value function, and can be constructed as shown in Equation 3 below. The value network 111 can be back propagated based on the time difference error, which is a cost function of the value network.

In the equation 2, V is a value function, w is a value network parameter, s _i is state information of the current episode i, S _{i +1} is state information of the next episode _{i + 1} , V _w (s _i ) is the compensation potential in the current episode, V _w (s _{i + 1} ) is the compensation potential in the next episode, and γ is the depreciation rate. At this time, r _{i + 1} can be configured to be received at the promo performance prediction artificial neural network device 10.

Accordingly, the value network can update the cost function of Equation (1) in the direction of gradient descent when the state of the promotion recommendation apparatus is changed.

According to the embodiment of the present invention, the Q value of the value network is supervised without starting from the random value while learning the value network separately from the policy network, so that the learning can be performed quickly. Accordingly, the burden of exploration in the action of selecting a combination of promotional properties having a very high complexity can be greatly reduced.

According to the promotional recommendation apparatus 11 according to the embodiment of the present invention, when the policy network 110 after completion of the guidance recommends the promotional attribute of the current episode i, when the value network 111 advances the recommended promotional attribute To be compensated. The policy network 110 and the value network 111 of the completed promotional recommendation apparatus 11 are utilized in combination with the simulation using the optimum promotion search module 112 to finally select a promotion attribute.

In addition, according to the value network 111 according to the embodiment of the present invention, an update of the policy network for outputting the probability of the recommended promotional attribute can be effected for each episode. In the existing reinforcement learning, there is a problem that the update of the reinforcement learning model is carried out after all the episodes have ended, so that it has been difficult to apply to the recommendation recommendation model.

The optimal promotional search module 112 performs a plurality of simulations for various states and various actions based on a plurality of agents calculated in a policy network and a value network to search for an optimal promotional attribute. The optimal promotional search module according to an embodiment of the present invention may utilize, for example, a Monte Carlo tree search, each node of the tree may be associated with a state, each edge may be associated with a particular action It represents the expected value (value) to follow, and it is a structure in which the leaf nodes are expanded each time a transition is made to a new state by taking a new action with the current state as a root node. In the optimal promotional search module according to an exemplary embodiment of the present invention, when the Monte Carlo tree search is utilized, the optimal promotional search can be processed in four steps of selection, expansion, evaluation, and backup.

The selection step of the optimal promotion search module 112 is a step of selecting the highest value action among the selectable actions until the leaf node comes out of the current state. In this case, we use the value of the value function stored at the edge and the visit frequency value to balance the inquiry-use balance. In the selection step, the formula for selecting an action is as follows.

In the above Equation 4 a _t is the action (performed promotion) at time t, Q (s _t, a) is the value of the value function is stored in the tree, u (s _t, a) is the state-of action pair It was used to balance exploration and utilization with inversely proportional to the number of visits.

The Expansion stage of the optimal promotion search module 112 is a step of adding a new node as a leaf node according to the probability of the policy network learned by the map learning when the simulation proceeds to the leaf node.

The evaluation step of the optimum promotional search module 112 is a step of evaluating the leaf node by using the value (compensation possibility) determined using the value network from the newly added leaf node and the compensation obtained from the leaf node to the end of the promotion episode using the policy network It is a step to evaluate value. The following equation is an example of evaluating the value of a new leaf node.

In Equation 5, V (s _L ) is the value of the leaf node, λ is the mixing parameter, v _θ (s _L ) is the value obtained through the value network, and z _L is the compensation obtained by continuing the simulation.

The backup step of the optimum promotion search module 112 is a step of re-evaluating the values of visited nodes and updating the visit frequency during the simulation by reflecting the value of the newly added leaf node. The following equation is an example of node value revaluation and visit frequency update.

In Equation (6), s _L ⁱ represents the leaf node in the i-th simulation, and 1 (s, a, i) represents the connection (s, a) visited in the i-th simulation. And to select the most visited connection (s, a) from the root node. According to the optimum promotional search module 112 according to the embodiment of the present invention, it is possible to select the optimal promotional attribute by preceding the simulation multiple times based on the value network for a plurality of promotional attributes selected by the policy network Is generated.

According to an embodiment of the present invention, the promotion recommendation apparatus 11 can be configured so that a plurality of agents (Agents) are configured. When a plurality of agents are configured, the promotions recommended by the promotional recommendation apparatus compete with each other for a specific state and a specific promotional attribute, so that the most optimal product and its promotion can be recommended within a certain budget.

5 is a flowchart illustrating an exemplary operation of the promotion recommendation apparatus 11 according to an embodiment of the present invention. 5, when the state s (t) is input by the Internet shopping mall data hub device 12 that tracks status data in a plurality of Internet shopping malls, a value (s) Various promotional properties are input to the optimal promotion search module 112 by agents and the recommended promotion attribute probability a (t), which is an action output by the optimum promotion search module 112, The episode t is ended and the episode t + 1 is started. In the episode t + 1, s (t + 1), which is a state change due to a (t), is input by the Internet shopping mall data hub device 12 and r (t + 1) So that the value network 111 and the policy network 110 are updated.

According to one embodiment of the present invention, in the case of the promotional image color information among the data included in the recommended promotion attribute data 500, according to the embodiment of the present invention, the actual promotional image is stored in the online shopping mall using artificial intelligence The promotional performance prediction and recommendation apparatus 1 can receive the corrected promotional image by modifying the color of the promotional image based on the color information of the promotional image. FIG. 6 is a schematic diagram showing an apparatus 13 for revising a promotion image according to an embodiment of the present invention. As shown in FIG. 6, the promotion image correcting apparatus 13 may be configured to include a generator and a ConvNet encoder.

The generator may be configured as an image generator composed of an encoder and a decoder such as a VAE and a GAN, and may receive the actual promotion image 210 and output the generated promoted image 212 with the changed color. The generated promotional image 212 is again received by the ConvNet encoder, and the ConvNet encoder outputs the encoded promotional image 211 and is cross entropy calculated with the promotional image color information of the recommended promotional attribute data 500 to output an error have. Such an error can be updated so that the generated promotional image 212 is fed back to the generator to further match the promotional image color information.

7 is a schematic diagram illustrating an example of a ConvNet encoder according to an embodiment of the present invention. As shown in Fig. 7, the ConvNet encoder can be constructed with [INPUT-CONV-RELU-POOL-FC]. In the case of the generated promoted image 212 as the input data, the input size is [32x32x3] when the INPUT input image has the width 32, the length 32, and the RGB channel. The CONV layer (Conv.Filter, 101) is connected to a part of the input image, and calculates the dot product of the weight and the connected area. The resulting volume will have the same size as [32x32x12]. The RELU layer is an activation function applied to each element such as max (0, x). The RELU layer does not change the size of the volume ([32x32x12]). As a result, Activation map 1 (102) is generated. The POOL layer 103 performs downsampling on the " horizontal and vertical " dimension to output a volume (Activation map 2, 104) such as [16x16x12]. A fully-connected (FC) layer 105 computes the class scores and outputs an output layer 106 of size [1x1x10]. &Quot; 10 " corresponds to a class score (promotion image color information according to an embodiment of the present invention) for 10 categories. The FC layer is connected to all the elements of the previous volume.

As such, ConvNet transforms the original image, which consists of pixel values, through each layer into a class score (promotional image color information according to one embodiment of the present invention). Some layers have parameters but some do not. In particular, the CONV / FC layers are activation functions that include not only the input volume, but also the weight and bias. RELU / POOL layers, on the other hand, are fixed functions. The parameters of the CONV / FC layer are learned in gradient descent so that the class score for each image is equal to the label of the image.

The parameters of the CONV layer consist of a series of learnable filters. Each filter is small in the horizontal / vertical dimension but covers the entire depth in the depth dimension. During the forward pass, each filter slides (precisely convolves) the horizontal and vertical dimensions of the input volume and creates a two-dimensional activation map. When sliding the filter over the input, a dot product is made between the filter and the input volume. In this way, ConvNet will learn the filter to activate for a specific pattern at a specific location in the input data. The accumulation of this activation map in depth dimension is the output volume. Therefore, each element of the output volume handles only a small area of the input, and neurons in the same activation map share the same parameters because they are the result of applying the same filter.

FIG. 8 is a flowchart illustrating a recommendation method according to an exemplary embodiment of the present invention. 8, the promotion recommendation method according to an embodiment of the present invention includes a promotion performance prediction step S10, a status information reception step S11, a recommended promotion attribute data output step S12, An information receiving step S13, a policy network and a value network updating step S14.

The promotional performance prediction step S10 is a step in which the promotional performance prediction ANN apparatus receives the existing promotion attribute data and performance data, predicts the promotion performance for the specific promotional attribute, and transmits the result to the promotion recommendation apparatus 11. Predicted promotional performance data can be used as an initial value for the policy and value networks.

The status information reception step S11 is a step in which the promotion recommendation apparatus 11 receives the status information s _t for the episode t in the Internet shopping mall data hub apparatus.

The recommended promotional attribute data output step S12 is a step in which the promotional recommendation apparatus 11 outputs the recommended promotional attribute data a _t including the recommended promotional attribute probability for the episode t through the optimum promotional search module 112 .

The state and compensation information receiving step S13 is a step in which the promotion recommendation apparatus 11 receives the state information s _{t + 1} for the episode t + 1 in the Internet shopping mall data hub device after the promotion is applied to the Internet shopping mall according to the recommended promotion attribute data ) And compensation information r _{t + 1} for a _t .

The policy network and value network updating step S14 is a step of updating the policy network and the value network based on the received state information s _{t + 1} and compensation information r _{t + 1} for a _t .

From the viewpoint of reinforcement learning, the objective of the promotion recommendation method according to the embodiment of the present invention is to improve the performance of the promotion, and the state can be divided into the present performance data (quantity, amount of sales, (Such as page view, conversion rate, average sales price, and sales profit), and environmental data (date, day, weather, current promotional products, shopping malls, Action may refer to promotion property data (promotion category (discount promotion, promotion promotion, etc.), promotion period, promotion shopping mall, exposure account, color of promotion image, etc.) May mean an increase or a decrease in performance data over a period of time.

As described above, those skilled in the art will appreciate that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. It is therefore to be understood that the above-described embodiments are to be considered in all respects only as illustrative and not restrictive. The scope of the present invention is defined by the appended claims rather than the detailed description, and all changes or modifications derived from the meaning and scope of the claims and their equivalents should be construed as being included in the scope of the present invention.

The features and advantages described herein are not all inclusive, and in particular, many additional features and advantages will be apparent to those skilled in the art in view of the drawings, specification, and claims. Moreover, it should be noted that the language used herein is primarily chosen for readability and for purposes of teaching, and may not be selected to delineate or limit the subject matter of the invention.

The foregoing description of embodiments of the invention has been presented for purposes of illustration. It is not intended to be exhaustive or to limit the invention to the precise form disclosed. Those skilled in the art will appreciate that many modifications and variations are possible in light of the above teachings.

The scope of the invention is, therefore, not to be limited by the Detailed Description, but is to be defined by the claims of any application based thereon. Accordingly, the disclosure of embodiments of the invention is illustrative and not restrictive of the scope of the invention, which is set forth in the following claims.

1: Promotion Performance Prediction and Recommendation Device in Online Shopping Mall Using Artificial Intelligence
10: Promotional Performance Predictive Artificial Neural Network Device
11: Promotional recommendation device
12: Internet shopping mall data hub device
13: Promotion image correction device
100: Product attribute data
110: Policy network
111: Value Network
112: Optimum Promotion Search Module
200: Promotion attribute data
300: Performance data
400: Promotion performance forecast data
500: Featured promotional property data

Claims (4)

A memory module for storing program codes of the promotional performance predictive artificial neural network module and the promotion recommendation module learned based on the promotion attribute data and the performance data of at least one online shopping mall; And
A processing module for processing the program code of the promotional performance artificial neural network module and the promotion recommendation module to predict the performance of the specific promotion and recommend the promotional attribute;
/ RTI >
Wherein the program code of the promotion recommendation module includes:
A promotional performance prediction step of receiving promotional performance prediction data, which is information on a promotional performance predicted for a specific promotional attribute in the promotional performance artificial neural network module;
A status information receiving step of receiving first status information (s _t ) for episode t;
Generating a compensation possibility information in a value network on the basis of the first state information and the promotion performance prediction data, and outputting a recommendation promotion property probability as a plurality of actions in a policy network based on the compensation possibility information Probability output step;
A recommended promotional attribute data output step of selecting and outputting recommended promotional attribute data (a _t ) for the episode t through an optimum promotional search based on the plurality of recommended promotional attribute probabilities;
The second state information s _{t + 1} for episode t + 1 and the compensation information r _{t + 1} for the recommended promotion attribute data a _t are received after the promotion based on the recommended promotional attribute data is applied And receiving compensation information; And
A policy network and a value network updating step of updating the policy network and the value network based on the received second state information s _{t + 1} and the compensation information r _{t + 1} ;
The computer program product being configured to run on a computer,
The first state information and the second state information may include performance data (such as a quantity, a sales amount, a page view, a conversion rate, an average selling price, And environmental data (date information, day of the week information, weather information, product information in progress, shopping mall information, or other product information of the shopping mall)
The recommended promotion attribute data means a promotion category, a promotion period, a promotional shopping mall, an exposed account or a color of a promotional image,
The compensation information means a rise or a fall of the performance data for a specific period,
The optimal promotional search in the step of outputting the recommended promotional attribute data is performed by using a Monte Carlo tree search. In the Monte Carlo tree search, each node of the tree represents a state, and each edge corresponds to a specific action It is composed of a structure in which a leaf node is expanded every time a transition is made to a new state by taking a new action with a current state as a root node,
The recommended promotional attribute data output step may include:
Selecting an action having the highest value among the selectable actions until the leaf node comes out of the current state;
An extension step of adding a new leaf node according to a probability of the policy network learned by the learning of the guidance when the simulation proceeds to the leaf node;
An evaluation step of evaluating the value of the new leaf node by calculating the compensation possibility information determined using the value network from the new leaf node and the compensation obtained from the leaf node until the end of the promotion episode using the policy network; And
An update step of re-evaluating the values of visited nodes during the simulation and updating the visit frequency by reflecting the value of the new leaf node;
Further comprising:
And selects the recommended promotional attribute data (a _t ) based on the visit frequency.
Promotion Performance Prediction and Recommendation Device in Online Shopping Mall Using Artificial Intelligence.
A promotional performance prediction module for receiving at least one promotional performance recommendation module for an online shopping mall, the promotional performance prediction data being information on predicted promotion performance for a specific promotional attribute in the promotional performance prediction artificial neural network module;
A status information receiving step in which the promotion recommendation module receives first status information (s _t ) for episode t;
The promotion recommendation module generates the compensation possibility information in the value network on the basis of the first state information and the promotion performance prediction data and then transmits the recommendation promotion property as a plurality of actions in the policy network based on the compensation possibility information An attribute probability output step of outputting a probability;
A recommended promotional attribute data output step of selecting and outputting recommended promotional attribute data (a _t ) for the episode t through an optimum promotional search based on the plurality of recommended promotional attribute probabilities;
After the promotion in accordance with the recommendation promotion attribute data is applied, the Promotions module episode t + 1 second state information (s _{t + 1)} and the like promotional attribute data (a _t) compensation information (r _t for about ₊₁ ), and receiving compensation information; And
A policy network and a value network updating step of updating the policy network and the value network based on the second state information (s _{t + 1} ) and the compensation information (r _{t + 1} ) received by the promotion recommendation module;
The computer program product being configured to run on a computer,
The status information includes performance data (such as quantity, sales amount, page view, conversion rate, average sales price, or sales profit) and environment data (date information, Day information, weather information, promotional product information, shopping mall information, or other product information of the shopping mall)
The recommended promotion attribute data means a promotion category, a promotion period, a promotional shopping mall, an exposed account or a color of a promotional image,
The compensation information means a rise or a fall of the performance data for a specific period,
The optimal promotional search in the step of outputting the recommended promotional attribute data is performed by using a Monte Carlo tree search. In the Monte Carlo tree search, each node of the tree represents a state, and each edge corresponds to a specific action It is composed of a structure in which a leaf node is expanded every time a transition is made to a new state by taking a new action with a current state as a root node,
The recommended promotional attribute data output step may include:
Selecting an action having the highest value among the selectable actions until the leaf node comes out of the current state;
An extension step of adding a new leaf node according to a probability of the policy network learned by the learning of the guidance when the simulation proceeds to the leaf node;
An evaluation step of evaluating the value of the new leaf node by calculating the compensation possibility information determined using the value network from the new leaf node and the compensation obtained from the leaf node until the end of the promotion episode using the policy network; And
An update step of re-evaluating the values of visited nodes during the simulation and updating the visit frequency by reflecting the value of the new leaf node;
Further comprising:
And selects the recommended promotional attribute data (a _t ) based on the visit frequency.
Promotion Performance Prediction and Recommendation Method in Online Shopping Mall Using Artificial Intelligence.
delete delete

Images