KR20190134967A - Promotional image improvement apparatus and method in online shopping mall using artificial intelligence - Google Patents

Abstract

The present invention relates to a promotion image modifying apparatus in an online shopping mall using artificial intelligence. According to one embodiment of the present invention, a promotion effect is increased by precisely modifying a color of a promotion image of an online shopping mall by learning of artificial intelligence.

Description

Promotional image improvement apparatus and method in online shopping mall using artificial intelligence}

The present invention relates to an apparatus for modifying a promotional image in an online shopping mall using artificial intelligence.

Promotions in online shopping malls can take the form of product discounts, sweepstakes, discount coupon payments and mileage payments. Since such promotions always cause a decrease in profits from the point of view of the online shopping mall, the type of promotion, the duration of the promotion, and the degree of promotion must be precisely designed for the persona. However, in the past, the design of the promotion was all empirically determined by the operator of the online shopping mall or the promotion agency.

In Korea, there are various types of online shopping such as open market / social commerce / general shopping mall / large mart shopping mall / card point mall, and most sellers sell multiple stores in the above online shopping malls. Difficulties arise in analyzing sales performance based on the data collected.

The components of the promotion are related to the date and conditions of the promotion and are causally related to the result of the promotion.However, various items such as season, month, day of the week, time, period, and holidays are included even when the date and time is one of the requirements It is very difficult for a person to calculate the contributions made by components based on the results.

In particular, in the case of the promotion depreciation rate that will benefit the consumer, the greater the depreciation rate, the greater the purchase benefit, which will increase the sales volume and increase sales. However, considering the cost of the product, it is important to balance the increase in sales volume and the appropriate rate of depreciation rate, and it is very difficult for a person to calculate whether the expected sales profit will be achieved. .

Web advertising recommendation method and system based on click pattern using a collaborative filtering system having a neural network, Korea Patent Registration 10-0792700, Naver Corporation

Accordingly, an object of the present invention in order to solve the low precision of the promotion of the online shopping mall that was previously empirically designed, in the online shopping mall using artificial intelligence, which predicts the performance of the promotion using artificial intelligence and recommends the characteristics of the promotion. To provide a promotion performance prediction and recommendation device and method.

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

An object of the present invention, the memory module for storing the program code of the artificial neural network that is already learned based on the product attribute data, promotion attribute data, other performance data and promotion; And a processing module for processing the program code of the artificial neural network to predict the performance of a specific promotion. The program code includes: an input step of receiving product attribute data (100), promotion attribute data (200), and performance data (300); A performance prediction step of generating promotion performance prediction data by predicting the performance of the specific promotion based on the product attribute data, the promotion attribute data and the performance data; And an output step of outputting the predicted promotion performance prediction data. The present invention may be achieved by providing an apparatus for predicting and recommending promotion performance in an online shopping mall using artificial intelligence.

In addition, the memory module for storing the program code of the promotion performance prediction artificial neural network module and the promotion recommendation module based on the promotion attribute data and the performance data; And a processing module for processing the program codes of the promotion performance prediction neural network module and the promotion recommendation module to predict the performance of a specific promotion and recommending promotion attributes, wherein the program code of the promotion recommendation module includes the promotion code. A promotion performance prediction step of receiving, in the performance prediction artificial neural network module, promotion performance prediction data which is information about the promotion performance predicted for a specific promotion attribute; Receiving status information s _t for episode t; After generating reward possibility information in a value network based on the status information and the promotion performance prediction data, an attribute probability output for outputting a recommended promotion attribute probability that is a plurality of actions in a policy network based on the reward probability information. step; A recommendation promotion attribute data output step of selecting and outputting recommendation promotion attribute data (a _t ) for the episode t through an optimal promotion search based on the plurality of recommendation promotion attribute probabilities; Receiving state and reward information for receiving status information s _{t + 1} for episode t + 1 and reward information r _{t + 1} for a _t after the promotion according to the recommended promotion attribute data is applied; And updating the policy network and the value network based on the received state information s _{t + 1} and the reward information r _{t + 1} . And the status information includes performance data (Quantity, Sales Amount, Page View, Conversion Rate, Average Sales Price, or Revenue revenue) and environmental data (date information, day of the week information, weather information, promotional product information, shopping mall information, or other product information of a shopping mall). By means of the exposure account or the promotional image color, wherein the reward information means the rise or fall of the performance data for a specific period, by providing a promotion performance prediction and recommendation device in the online shopping mall using artificial intelligence Can be achieved.

Another object of the present invention, the promotion recommendation module is a promotion performance prediction step of receiving promotion performance prediction data which is information on the promotion performance predicted for a specific promotion attribute in the promotion performance prediction artificial neural network module; Receiving status information by the promotion recommendation module for receiving status information s _t for episode t; After the promotion recommendation module generates reward possibility information in a value network based on the status information and the promotion performance prediction data, a recommendation promotion attribute probability of a plurality of actions in a policy network is generated based on the reward probability information. Outputting an attribute probability; A recommendation promotion attribute data output step of the promotion recommendation module selecting and outputting recommendation promotion attribute data (a _t ) for the episode t through an optimal promotion search based on the plurality of recommendation promotion attribute probabilities; Status and reward after the promotion according to the recommended promotion attribute data is applied, the promotion recommendation module receives status information s _{t + 1} for episode t + 1 and reward information r _{t + 1} for a _t Receiving information; And updating, by the promotion recommendation module, the policy network and the value network based on the received state information s _{t + 1} and the reward information r _{t + 1} . And the status information includes performance data (Quantity, Sales Amount, Page View, Conversion Rate, Average Sales Price, or Revenue revenue) and environmental data (date information, day of the week information, weather information, promotional product information, shopping mall information, or other product information of a shopping mall). By means of the exposure account or the promotional image color, wherein the reward information means the rise or fall of the performance data for a specific period, by providing a method for predicting and recommending promotion performance in the online shopping mall using artificial intelligence Can be achieved.

Another object of the present invention, a memory for storing the program code of the promotion recommendation module for recommending the promotion attribute based on the promotion attribute data and performance data for at least one online shopping mall and the promotion image modification module for modifying the color of the promotion image module; And a processing module for processing program codes of the promotion recommendation module and the promotion image modification module to recommend specific promotion attributes and to modify the color of the promotion image. The program code of the promotion recommendation module includes: receiving status information receiving first status information (s _t ) for episode t; Generating recommended reward attribute data based on the first state information, and outputting recommended promotion attribute data a _t for the episode t based on the reward possibility information; After the promotion according to the recommended promotion attribute data is applied, second state information s _{t + 1} for episode t + 1 and reward information r _{t + 1} for the recommended promotion attribute data a _t are received. Receiving state and reward information; And a weight updating step of updating a weight of the recommendation promotion attribute data outputting step based on the received second state information (s _{t + 1} ) and the reward information (r _{t + 1} ). It is configured to be performed on a computer including, wherein the program code of the promotional image modification module, Promotion image receiving step of receiving the promotional image; Generating a modified promotional image by modifying the color of the promotional image by an image generator including a VAE or a GAN; A comparison step of encoding the modified promotional image and calculating an error with recommended promotional image color information in the recommended promotional attribute data; And updating the image generator based on the error. The first status information and the second status information are configured to be performed on a computer, including performance data (Quantity, Sales Amount, Page View, Conversion Rate, and Average Sales). Average sales price or revenue) 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). It includes image color information, wherein the compensation information may be achieved by providing a promotional image modification apparatus in an online shopping mall using artificial intelligence, characterized in that it means the rise or fall of the performance data for a specific period.

It is another object of the present invention to provide a promotion recommendation module for recommending a promotion attribute based on promotion attribute data and performance data for at least one online shopping mall, and receiving status information about the episode t (s _t ). Receiving step; The promotion recommendation module generates reward possibility information based on the first state information, and then selects and outputs recommended promotion attribute data a _t for the episode t based on the reward possibility information; Data output step; After the promotion recommendation module applies the promotion according to the recommended promotion attribute data, the second state information s _{t + 1} for the episode t + 1 and the reward information r for the recommended promotion attribute data a _{t t + 1} ) receiving a state and reward information; And updating, by the promotion recommendation module, a weight of the recommendation promotion attribute data output step based on the received second state information (s _{t + 1} ) and the reward information (r _{t + 1} ). Promote image correction module for modifying the color of the promotional image, the promotion image receiving step of receiving the promotion image; Generating a modified promotional image by modifying the color of the promotional image by an image generator including a VAE or a GAN; A comparison step of encoding the modified promotional image and calculating an error with recommended promotional image color information in the recommended promotional attribute data; And updating the image generator based on the error. The first state information and the second state information include performance data (Quantity, Sales Amount, Inflow (Page View), Conversion Rate, Average Sales Price). Or sales revenue) and environmental data (date information, day of the week information, weather information, product information in progress of a promotion, shopping mall information, or other product information of a shopping mall), and the recommended promotion attribute data includes recommended promotion image color information. The reward information may be achieved by providing a method of modifying a promotional image in an online shopping mall using artificial intelligence, which means that the performance data is raised or lowered during a specific period of time.

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

First, according to an embodiment of the present invention, the promotion of the online shopping mall is elaborated and precisely made by the learning of artificial intelligence, thereby increasing the effect of the promotion.

Second, according to one embodiment of the present invention, the effect of the update of the policy network for outputting the probability of the recommended promotion attribute according to the value network 111 can proceed every episode. In the existing reinforcement learning, there is a problem that the update of the reinforcement learning model is performed after all episodes are finished, so it is difficult to apply to the promotion recommendation model.

Third, according to one embodiment of the present invention, the recommendation of the promotion is short-term and long-term targets, so even though there is difficulty in applying the existing DQN because the assumption of the MDP does not hold, the dependency of the policy network and the value network The promotion recommendation also has the effect of enabling reinforcement learning.

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

The following drawings, which are attached to this specification, illustrate exemplary embodiments of the present invention, and together with the detailed description thereof, serve to further understand the technical spirit of the present invention. It should not be interpreted.
1 is a view showing a promotion performance prediction and recommendation apparatus 1 in an online shopping mall using artificial intelligence according to an embodiment of the present invention,
Figure 2 is a schematic diagram showing a learning process of the promotion performance artificial neural network device according to an embodiment of the present invention,
3 is a schematic diagram showing an inference process of the promotion performance prediction artificial neural network device according to an embodiment of the present invention,
4 is a schematic diagram showing a promotion recommendation device 11 according to an embodiment of the present invention,
5 is a flowchart illustrating an operation example of the promotion recommendation device 11 according to an embodiment of the present invention;
6 is a schematic diagram showing a promotional image modification device 13 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;
8 is a flowchart illustrating a promotion recommendation method according to an embodiment of the present invention.

DETAILED DESCRIPTION Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, in describing in detail the principle of operation of the preferred embodiment of the present invention, if it is determined that the detailed description of the related known function or configuration may unnecessarily obscure the subject matter of the present invention, the detailed description thereof will be omitted.

In addition, the same reference numerals are used for parts having similar functions and functions throughout the drawings. Throughout the specification, when a particular part is connected to another part, this includes not only the case where it is directly connected, but also the case where it is indirectly connected with another element in between. In addition, the inclusion of a specific component does not exclude other components unless specifically stated otherwise, it means that may further include other components.

Apparatus and method for forecasting and recommending promotion performance in online shopping mall using artificial intelligence

1 is a block 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. As shown in FIG. 1, the promotion performance prediction and recommendation device in the online shopping mall using artificial intelligence according to an embodiment of the present invention is based on product attribute data, promotion attribute data, performance data, and promotions. It may include a performance prediction artificial neural network device 10 and the promotion recommendation device (11).

In relation to the promotion performance prediction neural network device 10, Figure 2 is a schematic diagram showing a learning process of the promotion performance prediction neural network device according to an embodiment of the present invention, Figure 3 is a promotion performance prediction according to an embodiment of the present invention It is a schematic diagram showing the reasoning process of artificial neural network device. As shown in Figures 2 and 3, the promotion performance prediction method in the online shopping mall using artificial intelligence receives the product attribute data 100, the promotion attribute data 200, the performance data 300, the promotion performance prediction artificial neural network The device 10 may be trained, and the product performance data 100 and the promotion property data 200 may be input to the pre-learned promotion performance prediction neural network device 10 to output the predicted promotion performance prediction data 400. have. In particular, when the promotion performance prediction neural network apparatus 10 learns, the weight of the hidden layer may be updated based on an error between the promotion performance data and the output data. In FIG. 2, each attribute constituting the product attribute data 100 and the promotion attribute data 200 is input to each node of an input layer such as x1, x2, and x3, and based on a weight such as w1, h1, h2, and h3. The promotional performance prediction data 400 predicted based on a cost function such as softmax after passing through a hidden layer, such as softmax, is output to the output layer of y1. Based on the prediction performance data and the error between the actual performance data (300) (error, -Sigma (y _i log p _i)), the weight of the promotion and prediction neural network apparatus 10 may be updated by back propagation.

Alternatively, a separate promotion performance prediction neural network device 10 may be configured for each product, and the promotion attribute data 200 and the performance data 300 may be received to learn the promotion performance prediction neural network device 10. The promotion attribute data 200 may be input to the pre-learned promotion performance prediction artificial neural network device to output the predicted promotion performance prediction data 400. In addition to the artificial neural network that can be used at this time, machine learning such as multi-layer perceptron, Naive-Bayesian Classification, Random Forest Classification can be used.

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

The promotion attribute data 200 may include a promotion category (discount promotion, free gift promotion, etc.), a promotion period, a promotion shopping mall, an exposure account, a promotion image, and the like. The promotional image data among the promotional attribute data according to an embodiment of the present invention may refer to a result value classified into a specific mood category and a specific color category through a previously learned CNN (Convolutional Neural Network).

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

In relation to the promotion attribute recommendation, the promotion recommendation apparatus 11 for outputting the recommendation promotion attribute data for outputting the highest promotion performance prediction data 400 using the promotion performance prediction artificial neural network device may be configured.

Promotion recommendation of the promotion recommendation device 11 according to an embodiment of the present invention may be performed using reinforcement learning. 4 is a schematic diagram showing a promotion recommendation device 11 according to an embodiment of the present invention. As shown in Figure 4, the promotion recommendation device 11 according to an embodiment of the present invention is the probability of the value network 111 and the recommended promotional attributes that are artificial neural network learning a value function that outputs the value in a specific state It may include a policy network 110 for learning a policy function to output, the policy network 110 and the value network 111 according to an embodiment of the present invention is connected to the promotion performance prediction artificial neural network device 10 It may be configured to. 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 terms of reinforcement learning, the objective of the promotion recommendation device 11 according to an embodiment of the present invention is to improve the performance of the promotion, and the state is the current performance data (Quantity, Sales Amount). ), Page View, Conversion Rate, Average Sales Price, Revenue, etc., and environmental data (date, day of the week, weather, current promotional products, shopping malls, other product information in the mall) ) May mean promotional attribute data (promotion category (discount promotions, free gift promotions, etc.), promotional periods, promotional stores, impression accounts, promotional image colors, etc.), and rewards. ) Can mean the rise or fall of performance data over a period of time.

The policy network 110 is an artificial neural network that determines the probability of specific attributes of the recommended promotion in each state of the promotion recommendation device 11, and outputs the recommended promotion attribute probability by learning a policy function. The cost function of the policy network may be a function that calculates cross entropy by multiplying the policy function and the cost function of the value network, and then takes a policy gradient. For example, the policy function may be configured as shown in Equation 1 below. The policy network can be back propagated based on the product of cross entropy and time difference error, which is a cost function of the value chain.

In Equation 1, π is a policy function, θ is a policy network parameter, π _θ (a _i s _i ) is the likelihood that a particular action (promotion of a particular property) will be performed in the current episode, V is a value function, and w is a value chain. Parameter, s _i is the status information of the current episode i, S _{i + 1} is the status information of the next episode i + 1, r _{i + 1} is the reward expected to be obtained in the next episode, and V _w (s _i ) is The reward potential in the current episode, V _w (s _{i + 1} ) may be the reward potential in the next episode, and γ may mean a depreciation rate. In this case, r _{i + 1} may be configured to receive from the promotion performance prediction neural network device 10. As a result, the policy network 110 according to an embodiment of the present invention initially outputs a promotion attribute that simulates the promotion history of the user through a policy gradient.

The policy network 110 according to an embodiment of the present invention is supervised learning based on existing promotion attribute data and performance data before reinforcement learning is performed, and thus the weight of the policy network is updated. You can learn the basics. In other words, the weight of the policy network may be supervised and set based on existing promotion attribute data and performance data. According to this, there is an effect that the policy network can be learned very quickly by recording the existing promotion.

In addition, according to an embodiment of the present invention may be configured to be supervised learning on the basis of the existing promotion attribute data and the resulting performance data including a random vector during the supervised learning of the policy network 110. The random vector may use, for example, a Gaussian distribution. According to this, there is an effect that the policy network can output a challenging promotion policy with a random probability. When the supervised learning of the policy network 110 is configured to be supervised learning based on the existing promotion attribute data and the performance data, the recommendation of the promotion is optimized within the existing policy. However, if a random vector is included in the supervised learning of the policy network according to an embodiment of the present invention, as the reinforcement learning proceeds, the policy network can learn more effective promotion than the existing policy.

The value network 111 is an artificial neural network that derives the possibility of achieving a reward in each state that the promotion recommendation device 11 may have, and learns a value function. The value network 111 presents directions for which direction the promotion recommendation device 11, which is an agent, is updated. To this end, the input variable of the value network 111 is set to state information that is information about the state of the promotion recommendation device 11, the output variable of the value network 111 is the promotion recommendation device 11 to achieve the reward It may be set as reward possibility information that is a possibility. Compensation probability information according to an embodiment of the present invention may be calculated by a Q-function as shown in the following equation.

In Equation 2, Q _π may indicate overall reward possibility information in the future when state s and action a in a specific policy π, R may mean compensation for a specific period, and gamma may mean a depreciation rate. S _t may represent a state of time t, A _t may represent an action of time t, and E may represent an expected value. Compensation possibility information (Q value) according to an embodiment of the present invention defines the update direction and size of the policy network 110.

In this case, the cost function of the value network may be a mean square error (MSE) function for the value function, and may be configured as shown in Equation 3 below. The value chain 111 may be back propagated based on a time difference error that is a cost function of the value chain.

In Equation 2, V is a value function, w is a value chain parameter, s _i is status information of the current episode i, S _{i + 1} is status information of the next episode _{i + 1} , and r _{i + 1} is The reward expected to be obtained, V _w (s _i ) may be the reward potential in the current episode, V _w (s _{i + 1} ) may be the reward probability in the next episode, and γ may mean a depreciation rate. In this case, r _{i + 1} may be configured to receive from the promotion performance prediction neural network device 10.

Accordingly, the value network may be updated in a direction in which the cost function of Equation 1 is gradientd when the state of the promotion recommendation device is changed.

According to an embodiment of the present invention, while learning the value network separately from the policy network, the Q value of the value network is supervised instead of starting at random, so that the rapid learning is possible. This results in an effect of greatly reducing the exploration burden in the action of selecting a combination of highly complex promotional attributes.

According to the promotion recommendation device 11 according to an embodiment of the present invention, when the policy network 110 that has completed the map learning recommends the promotion attribute of the current episode i, the value network 111 proceeds to the recommended promotion attribute. Is learned to predict the reward. The policy network 110 and the value network 111 of the promotion recommendation device 11 that have completed the learning are combined with a simulation using the optimal promotion search module 112 and finally used to select promotion attributes.

In addition, according to the value network 111 according to an embodiment of the present invention, there is an effect that the update of the policy network that outputs the probability of the recommended promotion attribute may proceed every episode. In the existing reinforcement learning, there is a problem that the update of the reinforcement learning model is performed after all episodes are finished, so it is difficult to apply to the promotion recommendation model.

The optimal promotion search module 112 is configured to search for an optimal promotion attribute by conducting 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. Optimal promotion search module according to an embodiment of the present invention, for example, may utilize Monte Carlo tree search, each node of the tree to a state, each edge to a specific action for that state It represents the expected value, and the leaf node expands whenever the new state is taken by taking the current state as the root node and taking a new action. In the optimal promotion search module according to an embodiment of the present invention, the optimal promotion search may be processed in four stages of selection, expansion, evaluation, and backup when the Monte Carlo tree search is utilized.

The selection step of the optimal promotion search module 112 is a step of selecting and selecting the most valuable action from the current state until the leaf node emerges from the current state. At this time, the value of the value function stored in the edge is used and the visit frequency value to balance the exploration-use. The equation for action selection in the selection step is shown below.

In Equation 4, a _t is an action at time t (promotion is performed), Q (s _t , a) is a value function stored in the tree, and u (s _t , a) is the state-action pair. Inversely proportional to the number of visits, it is used to balance exploration and usage.

The expansion step of the optimum promotion search module 112 is a step of adding a new node to the leaf node by acting according to the probability of the policy network learned through supervised learning when the simulation proceeds to the leaf node.

The evaluation stage of the optimal promotion search module 112 uses the value chain determined from the newly added leaf node (reward potential) and the reward obtained by proceeding until the end of the promotion episode using the policy network from the leaf node. It is a step in evaluating value. The following equation is an example of evaluating the value of a new leaf node.

In Equation 5, V (s _L ) may be a leaf node value, λ is a mixing parameter, v _θ (s _L ) may be a value obtained through a value network, and z _L may be a reward obtained by continuing a simulation.

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

In Equation 6 above, s _L ⁱ represents a leaf node in the i-th simulation, 1 (s, a, i) represents whether a visit (s, a) is visited in the i-th simulation, and when the tree search is completed, the algorithm It may be configured to select the most visited connection (s, a) from the root node. According to the optimum promotion search module 112 according to an embodiment of the present invention, the effect of being able to select the optimal promotion attribute by preceding the simulation multiple times based on the value network for the plurality of promotion attributes selected by the policy network. Is generated.

According to an embodiment of the present invention, the promotion recommendation device 11 may be configured to configure a plurality of agents. When a plurality of agents are configured, promotions recommended by the promotion recommendation device for each of a specific state and a specific promotion attribute may compete with each other, and thus, the most optimal product and a promotion thereof may be recommended within a predetermined budget.

5 is a flowchart illustrating an operation example of the promotion recommendation device 11 according to an embodiment of the present invention. As shown in FIG. 5, when the state s (t) is input by the Internet shopping mall data hub device 12 that tracks the state data in the plurality of internet shopping malls, the plurality of policy networks 110 are determined by the value network 111. Various promotion attributes are inputted to the optimal promotion search module 112 by two agents, and the promotion is performed by the recommended promotion attribute probability a (t) which is an action output by the optimal promotion search module 112. This progression ends episode t and episode t + 1 begins. In episode t + 1, s (t + 1), which is a state change caused by a (t), is again input by the Internet shopping mall data hub device 12, and r (t + 1), which is a reward according to a (t), It is input immediately to update the value network 111 and policy network 110.

In addition, according to an 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 an embodiment of the present invention in the online shopping mall using artificial intelligence The promotion performance prediction and recommendation device 1 may receive the modified promotion image by correcting the color of the promotion image based on the promotion image color information. 6 is a schematic diagram showing a promotional image modification device 13 according to an embodiment of the present invention. As shown in FIG. 6, the promotional image modification device 13 may be configured to include a generator and a ConvNet encoder.

The generator may be configured as an image generator including encoders and decoders such as VAE, GAN, etc., and may receive the actual promotional image 210 and output the generated promotional image 212 having changed color. The generated promotional image 212 is received by the ConvNet encoder again, and the ConvNet encoder outputs the encoded promotional image 211 to calculate cross-entropy with the promotional image color information of the recommended promotional attribute data 500 and output an error. have. This error may be updated so that the generated promotional image 212 is fed back to the generator to generate more for the promotional image color information.

7 is a schematic diagram showing an example of a ConvNet encoder according to an embodiment of the present invention. As shown in FIG. 7, the ConvNet encoder may be constructed with [INPUT-CONV-RELU-POOL-FC]. In the case of the generated promotional image 212 that is input data, when the INPUT input image has 32 horizontal, 32 vertical, and RGB channels, the size of the input is [32x32x3]. The CONV layer (Conv. Filter) 101 is connected to some areas of the input image and calculates a dot product of the connected areas and their weights. The resulting volume will be 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, we create Activation map 1 (102). The POOL layer (pooling) 103 performs downsampling on the "horizontal and vertical" dimension and outputs a reduced volume (Activation map 2, 104) such as [16x16x12]. The fully-connected layer 105 calculates class scores and outputs a volume 106 having a size of [1 × 1 × 10]. "10" corresponds to a class score (promotional image color information according to an embodiment of the present invention) for ten categories. The FC layer is connected to all the elements of the previous volume.

As such, ConvNet transforms the original image consisting of pixel values into class scores (promotional image color information according to an embodiment of the present invention) through each layer. Some layers have parameters, while others do not have parameters. In particular, CONV / FC layers are activation functions that include weight and bias as well as input volume. 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 that image.

The parameters of the CONV layer consist of a set of learnable filters. Each filter is small in the horizontal / vertical dimension but encompasses the full depth in the depth dimension. In the forward pass, each filter slides (or precisely convolves) each filter into the horizontal / vertical dimension of the input volume and generates a two-dimensional activation map. When sliding the filter over the input, a dot product occurs between the filter and the input volume. This process allows ConvNet to learn filters that activate specific patterns at specific locations in the input data. Stacking these activation maps in the 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.

8 is a flowchart illustrating a promotion recommendation method according to an embodiment of the present invention. As shown in Figure 8, the promotion recommendation method according to an embodiment of the present invention, the promotion performance prediction step (S10), the status information receiving step (S11), the recommended promotion attribute data output step (S12), status and reward It may include the information receiving step (S13), policy network and value network update step (S14).

The promotion performance prediction step (S10) is a step in which the promotion performance prediction artificial neural network device receives the existing promotion attribute data and the performance data, predicts the promotion performance for the specific promotion attribute, and transmits it to the promotion recommendation apparatus 11. The predicted promotional performance data can be used as an initial value for the policy and value chains.

In the state information receiving step S11, the promotion recommendation device 11 receives the state information s _t for the episode t from the Internet shopping mall data hub device.

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

In the step S13 of receiving status and reward information, after the promotion is applied to the Internet shopping mall according to the recommended promotion property data, the promotion recommendation device 11 performs status information about episode t + 1 in the Internet shopping mall data hub device (s _{t + 1).} ) And the 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 reward information r _{t + 1} for a _t .

In terms of reinforcement learning, the objective of the promotion recommendation method according to an embodiment of the present invention is to improve the performance of the promotion, and the state is the current performance data (Quantity, Sales Amount, Inflow). (Page View), Conversion Rate, Average Sales Price, Revenue, etc.) and environmental data (date, day of the week, weather, current promotional products, shopping malls, and other product information). Action may mean promotion attribute data (promotion category (discount promotion, free gift promotion, etc.), promotion period, promotion store, exposure account, promotional image color, etc.), and reward may be It can mean the rise or fall of performance data over time.

As described above, those skilled in the art will appreciate that the present invention can be implemented in other specific forms without changing the technical spirit or essential features. Therefore, the above-described embodiments are to be understood in all respects as illustrative and not restrictive. The scope of the present invention is shown by the following claims rather than the detailed description, and all changes or modifications derived from the meaning and scope of the claims and equivalent concepts should be construed as being included in the scope of the present invention.

The features and advantages described herein do not include all, and in particular many additional features and advantages will become 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 has been chosen primarily for ease of reading and for teaching purposes, and may not be selected to describe or limit the subject matter of the present invention.

The foregoing description of the 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 can appreciate that many modifications and variations are possible in light of the above disclosure.

Therefore, the scope of the present invention is not limited by the detailed description, but is defined by any claims of the application on which it is based. Accordingly, the disclosure of the embodiments of the present invention is illustrative and does not limit the scope of the invention described in the claims below.

1: Predicting and recommending promotion performance in online shopping mall using artificial intelligence
10: Promotion performance forecast neural network device
11: promotional recommendation device
12: Internet shopping mall data hub device
13: promotional image correction device
100: product attribute data
110: policy network
111: value chain
112: Optimal promotion navigation module
200: Promotional Attribute Data
300: performance data
400: Promotional performance forecast data
500: Referral promotion attribute data

Claims (1)

The program code of the promotion performance prediction neural network module, the promotion recommendation module recommending the promotion attribute, and the promotion image modification module to modify the color of the promotion image based on the promotion attribute data and the performance data of the at least one online shopping mall. A memory module for storing; And
A processing module for processing program codes of the promotion performance prediction artificial neural network module, the promotion recommendation module, and the promotion image modification module to recommend specific promotion attributes and to modify the color of the promotion image;
Including,
The program code of the promotion recommendation module,
A promotion performance prediction step of receiving, at the promotion performance prediction artificial neural network module, promotion performance prediction data which is information on the promotion performance predicted for a specific promotion attribute;
Receiving status information receiving first status information s _t for the episode t;
An attribute for generating reward possibility information in a value network based on the first state information and the promotion performance prediction data, and then outputting a recommended promotion attribute probability that is a plurality of actions in a policy network based on the reward probability information; Probability output step;
A recommendation promotion attribute data output step of selecting and outputting recommendation promotion attribute data (a _t ) for the episode t through an optimal promotion search based on the plurality of recommendation promotion attribute probabilities;
After the promotion according to the recommended promotion attribute data is applied, second state information s _{t + 1} for episode t + 1 and reward information r _{t + 1} for the recommended promotion attribute data a _t are received. Receiving state and reward information; And
A policy network and 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 reward information r _{t + 1} ;
Is configured to run on a computer, including
The program code of the promotional image modification module,
A promotional image receiving step of receiving the promotional image;
Generating a modified promotional image by modifying the color of the promotional image by an image generator including a VAE or a GAN;
A comparison step of encoding the modified promotional image and calculating an error with recommended promotional image color information in the recommended promotional attribute data; And
An update step of updating the image generator based on the error;
It is configured to run on a computer, including
The first state information and the second state information are performance data (Quantity, Sales Amount, Inflow (Page View), Conversion Rate, Average Sales Price, or Sales Profit). And environmental 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 includes recommended promotion image color information.
The reward information refers to the rise or fall of the performance data for a specific period of time,
The policy network is an artificial neural network for outputting the recommended promotion attribute probability in each state, and the cost function of the policy network is configured as in Equation 1 below.
The value network is an artificial neural network for generating the rewardability information that is the possibility of achieving a reward in each state, the cost function of the value network is characterized in that it is configured as
Device for modifying promotion image in online shopping mall using policy network and value network:
[Equation 1]

[Equation 2]

In Equation 1, π is a policy function, θ is the possibility to network policy parameters, π _{θ (i} a │s _i) has a specific action in the current episode (promotion of a particular property), V value is a function, w is a value Network parameter, s _i is the status information of the current episode i, S _{i + 1} is the status information of the next episode i + 1, r _{i + 1} is the reward expected to be obtained in the next episode, V _w (s _i ) Is the reward probability in the current episode, V _w (s _{i + 1} ) is the reward probability in the next episode, and γ is the depreciation rate.
In Equation 2, V is a value function, w is a value network parameter, s _i is the status information of the current episode i, S _{i + 1} is the next episode i + 1 status information, r _{i + 1} is the next episode The reward expected to be obtained from V _w (s _i ) is the reward probability in the current episode, V _w (s _{i + 1} ) is the reward probability in the next episode, and γ is the depreciation rate.

Images