CN111242654B - Method and system for generating advertisement picture - Google Patents
Method and system for generating advertisement picture Download PDFInfo
- Publication number
- CN111242654B CN111242654B CN201811432961.4A CN201811432961A CN111242654B CN 111242654 B CN111242654 B CN 111242654B CN 201811432961 A CN201811432961 A CN 201811432961A CN 111242654 B CN111242654 B CN 111242654B
- Authority
- CN
- China
- Prior art keywords
- picture
- style
- state
- user
- selecting
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
- G06Q30/0269—Targeted advertisements based on user profile or attribute
- G06Q30/0271—Personalized advertisement
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Economics (AREA)
- Marketing (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The disclosure provides a method and a system for generating an advertisement picture, and relates to the technical field of picture processing. The method comprises the following steps: carrying out picture processing on the picture of the commodity to obtain a main picture of the commodity; obtaining a picture style corresponding to the user behavior according to the user behavior log; selecting a material space according to the style of the picture; selecting, from a plurality of material spaces, a material that maximizes a function value of a cost function in a corresponding picture state, wherein arguments of the cost function include the picture state and an action to select the material from the corresponding material space, the picture state representing a selected combination of materials; and generating an advertisement picture of the commodity according to the selected material. The advertisement pictures can be generated according to the favorite style of the user, the requirements of individuation and aesthetic dynamism of the user are met, and therefore the user experience is improved.
Description
Technical Field
The present disclosure relates to the field of image processing technologies, and in particular, to a method and a system for generating an advertisement image.
Background
With the continuous increase of interest of electronic commerce platform users on commodity quality, in order to provide better user experience to attract more users to visit and promote the increase of platform transaction volume, the pages of the electronic commerce platform increasingly use various exquisite commodity pictures and texts to attract the users. The commodities which are interesting to customers are changed day by day, and tens of thousands of advertisement commodity pictures and texts displayed in front of the users are naturally updated every day on the page of the E-commerce platform, which relates to a large amount of picture design work. It is impractical to have a designer perform such a large amount of work, with the enormous amount of effort. And the manual efficiency cannot keep up with the speed of the demand update. Therefore, in the related art, a picture can be automatically generated according to a predetermined rule.
Disclosure of Invention
The inventors of the present disclosure found that the technical solutions in the related art produce almost the same pictures for the same kind of goods due to the inflexibility and limitation of the rules themselves. This is contrary to the need for customers to see different novel merchandise pictures. Moreover, the picture seen by each customer is the same, so that some customers do not like the picture, and the aesthetic difference of different customers is not considered.
In view of this, the technical problem that this disclosure solved is: the method for generating the advertisement picture is improved, and the advertisement picture can be generated according to the favorite style of the user.
According to an aspect of the embodiments of the present disclosure, there is provided a method of generating an advertisement picture, including: carrying out picture processing on a picture of a commodity to obtain a main picture of the commodity; obtaining a picture style corresponding to the user behavior according to the user behavior log; selecting a material space according to the picture style; selecting, from a plurality of material spaces, a material that maximizes a function value of a cost function in a respective picture state, wherein arguments of the cost function include a picture state representing a selected combination of materials and an action to select the material from the corresponding material space; and generating an advertisement picture of the commodity according to the selected material.
In some embodiments, the step of selecting from the plurality of material spaces the material that maximizes the function value of the cost function in the respective picture state comprises: selecting a material which enables the function value of the value function to be maximum under the current picture state from the current material space; adding the currently selected materials into the selected material combination to obtain the next picture state; and selecting a material from the next material space that maximizes the function value of the cost function in the next picture state.
In some embodiments, prior to selecting material from the plurality of material spaces, the method further comprises: constructing the cost function.
In some embodiments, the step of obtaining a picture style corresponding to the user behavior from the user behavior log comprises: predicting a picture style corresponding to the user behavior according to the user behavior log by using a style generation model; in the model training stage, a picture with a marked picture style is input into the style generation model, the style generation model calculates a style prediction result of the picture, and the style prediction result and the style true result are compared to update parameters of the style generation model.
In some embodiments, where material is not selected in the initial stage, the state of the currently selected combination of material is a 0 vector, and the initial value of the cost function is 0.
In some embodiments, the material space comprises: decorative patterns, layouts, color combinations, and background textures.
In some embodiments, the user behavior log includes pictures of merchandise that the user browses, clicks, purchases, or collects.
In some embodiments the method further comprises: and updating the value function by utilizing a depth certainty strategy gradient algorithm in response to the feedback operation of the user on the advertisement picture.
In some embodiments, the method further comprises: and collecting the pictures clicked by the user to update the favorite picture style of the user.
According to another aspect of the embodiments of the present disclosure, there is provided a system for generating an advertisement picture, including: the picture processing unit is used for carrying out picture processing on the pictures of the commodities to obtain main pictures of the commodities; the style acquisition unit is used for acquiring the style of the picture corresponding to the user behavior according to the user behavior log; the material space selection unit is used for selecting a material space according to the picture style; a material selection unit configured to select, from a plurality of material spaces, a material that maximizes a function value of a cost function in a corresponding picture state, wherein the independent variables of the cost function include a picture state representing a combination of the selected materials and an action of selecting the material from the corresponding material space; and an advertisement picture generating unit for generating an advertisement picture of the commodity according to the selected material.
In some embodiments, the material selection unit is configured to select, from a current material space, a material that maximizes a function value of the cost function in a current picture state; adding the currently selected materials into the selected material combination to obtain the next picture state; and selecting a material from the next material space that maximizes the function value of the cost function in the next picture state.
In some embodiments, the system further comprises: a function construction unit for constructing the cost function.
In some embodiments, the style obtaining unit is configured to predict, according to the user behavior log, a picture style corresponding to the user behavior by using a style generation model; in the model training stage, a picture with a marked picture style is input into the style generation model, the style generation model calculates a style prediction result of the picture, and the style prediction result and the style true result are compared to update parameters of the style generation model.
In some embodiments, where material is not selected in the initial stage, the state of the currently selected combination of material is a 0 vector, and the initial value of the cost function is 0.
In some embodiments, the material space comprises: decorative patterns, layouts, color combinations, and background textures.
In some embodiments, the user behavior log includes pictures of merchandise that the user browses, clicks, purchases, or collects.
In some embodiments, the system further comprises: and the updating unit is used for responding to the feedback operation of the user on the advertisement picture and updating the value function by utilizing a depth certainty strategy gradient algorithm.
In some embodiments, the style acquiring unit is further configured to collect pictures clicked by the user to update the style of the pictures liked by the user.
According to another aspect of the embodiments of the present disclosure, there is provided a system for generating an advertisement picture, including: a memory; and a processor coupled to the memory, the processor configured to perform the method as previously described based on instructions stored in the memory.
According to another aspect of embodiments of the present disclosure, there is provided a computer readable storage medium having stored thereon computer program instructions which, when executed by a processor, implement the steps of the method as previously described.
In the method, the picture of the commodity is subjected to picture processing to obtain a main picture of the commodity. And obtaining the picture style corresponding to the user behavior according to the user behavior log. A material space is selected according to the picture style. The material that maximizes the function value of the cost function in the corresponding picture state is selected from the plurality of material spaces. The arguments of the cost function include the picture state and the action of selecting material from the corresponding material space. The picture status represents the selected combination of material. And generating an advertisement picture of the commodity according to the selected material. The method can generate the advertisement pictures according to the favorite style of the user, and meets the requirements of user individuation and aesthetic dynamism, thereby improving the user experience.
Other features of the present disclosure and advantages thereof will become apparent from the following detailed description of exemplary embodiments thereof, which proceeds with reference to the accompanying drawings.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the disclosure and together with the description, serve to explain the principles of the disclosure.
The present disclosure may be more clearly understood from the following detailed description, taken with reference to the accompanying drawings, in which:
FIG. 1 is a flow diagram illustrating a method of generating an advertising picture in accordance with some embodiments of the present disclosure;
FIG. 2 is a flow diagram illustrating a method of generating advertisement pictures according to further embodiments of the present disclosure;
FIG. 3 is a flow diagram illustrating a depth deterministic policy gradient algorithm according to some embodiments of the present disclosure;
FIG. 4 is a block diagram that schematically illustrates a system for generating advertising pictures, in accordance with some embodiments of the present disclosure;
FIG. 5 is a block diagram that schematically illustrates a system for generating advertisement pictures, in accordance with further embodiments of the present disclosure;
FIG. 6 is a block diagram that schematically illustrates a system for generating advertisement pictures, in accordance with further embodiments of the present disclosure;
FIG. 7 is a block diagram that schematically illustrates a system for generating advertisement pictures, in accordance with further embodiments of the present disclosure;
FIG. 8 is a block diagram that schematically illustrates a system for generating advertisement pictures, in accordance with further embodiments of the present disclosure.
Detailed Description
Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that: the relative arrangement of the components and steps, the numerical expressions, and numerical values set forth in these embodiments do not limit the scope of the present disclosure unless specifically stated otherwise.
Meanwhile, it should be understood that the sizes of the respective portions shown in the drawings are not drawn in an actual proportional relationship for the convenience of description.
The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the disclosure, its application, or uses.
Techniques, methods, and apparatus known to those of ordinary skill in the relevant art may not be discussed in detail but are intended to be part of the specification where appropriate.
In all examples shown and discussed herein, any particular value should be construed as merely illustrative, and not limiting. Thus, other examples of the exemplary embodiments may have different values.
It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, further discussion thereof is not required in subsequent figures.
Fig. 1 is a flow chart illustrating a method of generating an advertisement picture according to some embodiments of the present disclosure. As shown in fig. 1, the method includes steps S102 to S110.
In step S102, a picture of the product is subjected to picture processing to obtain a main picture of the product.
For example, an image processing method such as adaboost can be used to perform body matting on a picture of a to-be-recommended commodity, so as to obtain a body picture of the commodity.
In step S104, a picture style corresponding to the user behavior is obtained from the user behavior log.
It should be noted that "user" described herein may be a single user or a group user. The style of a picture refers to the style type to which the picture belongs. For example, the picture style may include a decoration pattern, a layout, a color combination, a background texture, and the like, which are involved in the picture.
In some embodiments, the user behavior log may include pictures of merchandise that the user browses, clicks, purchases, or collects. Here, browsing, clicking, purchasing or collecting, etc. are user behaviors.
In addition, the picture style corresponding to the user behavior refers to a picture style preferred by the user corresponding to the overall expression of the user behavior. For example, if a more a-type picture style of the user is obtained from the overall expression of the user's browsing, clicking, purchasing, or collecting behavior, the picture style corresponding to the user behavior may be obtained as the a-type picture style.
In some embodiments, this step S104 may include: and predicting the picture style corresponding to the user behavior according to the user behavior log by using a style generation model. For example, the style generation model is a convolutional neural network model. In the model training stage, the picture with the marked picture style is input into the style generation model, the style generation model calculates to obtain a style prediction result of the picture, and the style prediction result is compared with the style reality result to update the parameters of the style generation model. For example, the model parameters may be updated by a reverberation propagation error technique, so that the prediction accuracy of the style generation model to the picture style is higher and higher. For example, the style prediction result is a prediction function value obtained by the style generation model calculating the style of the picture using the function used, and the style reality result is a reality function value corresponding to the style of the picture.
After the model is trained, advertisement pictures clicked by the user in the user behavior log are input into the style generation model, and the style mainly represented by the pictures is output and serves as the style preferred by the user.
In step S106, a material space is selected according to the picture style.
In some embodiments, the material space may include: decorative patterns, layouts, color combinations, and background textures. For example, a material space for decoration patterns, layout ways, color combinations, and background textures is selected according to the picture style.
For example, there are four material spaces according to the four stages of generating a picture: decorative patterns, layouts, color combinations, and background textures. These four material spaces are subsets of the four very large material libraries that are screened. Each material is machine learning trained to apply style labels when added to the material library. And when a style is determined, classifying each material in the material library by using a decision tree to obtain a material space.
In step S108, a material that maximizes the function value of the cost function in the corresponding picture state is selected from the plurality of material spaces. The arguments of the cost function may include the picture state and the action of selecting material from the corresponding material space. The picture status represents the selected combination of material. That is, the cost function may include two arguments: status and action. The state represents a mathematical description of the combination of materials that have been selected and the action represents the material selected at the current stage. For example, the material that maximizes the function value of the cost function in the corresponding picture state may be selected from a plurality of material spaces in sequence.
In some embodiments, this step S108 may include: selecting a material which enables the function value of the value function to be maximum under the current picture state from the current material space; adding the currently selected materials into the selected material combination to obtain the next picture state; and selecting the material which enables the value function of the value function to be maximum in the next picture state from the next material space. This is until one material is selected from each material space.
For example, the picture of the commodity subject obtained in step S102 may be input to the picture generator. A cost function is constructed that describes each state and selected material. When the material is selected in each stage, the material with the largest value of the value function of the current state can be selected. Thus, one material is selected in the material space of the current stage by one step at each stage according to the picture generation strategy.
For example, the cost function can be represented by Q (s, a), where s is the state (i.e., the state of the selected material combination), a is the action of selecting a material from the corresponding material space (e.g., the action of selecting a material from the current material space), and Q is the cumulative value obtained by the action of selecting a material at a certain state s. Cumulative value refers to the sum of all states that receive immediate revenue from the beginning of a state to the end of the state.
For example, the cost function Q under policy ππ(s, a) is:
Qπ(s,a)＝Eπ[Rt+1+γQπ(St+1,At+1)|St＝s,At＝a] (1)
wherein Q represents a cost function, S is a state, A is an action, Rt+1For immediate revenue, π is the policy. The lower case s represents a certain state determined, and the lower case a represents an action of selecting a certain material from the material space (i.e., a certain action determined). So QπThe value of (s, a) is equal to the expectation of the cumulative benefit to be obtained on the premise that (s, a) occurs. QπIn (s, a), s and a are known for certainty.
In some embodiments, where material is not selected in the initial stage, the state of the currently selected combination of material is a 0 vector, and the initial value of the cost function is 0.
In step S110, an advertisement picture of the product is generated from the selected material. For example, each time a material is selected, the material is directly filled in according to the position of the material in the picture. When the last material is selected, the picture can be generated.
To this end, methods of generating advertisement pictures according to some embodiments of the present disclosure are provided. In the method, picture processing is performed on a picture of a commodity to obtain a main picture of the commodity. Obtaining a picture style corresponding to the user behavior according to the user behavior log; a material space is selected according to the picture style. Selecting a material from the plurality of material spaces that maximizes the function value of the cost function in the corresponding picture state. The arguments of the cost function include the picture state representing the combination of materials that have been selected and the action of selecting materials from the corresponding material space. And generating an advertisement picture of the commodity according to the selected material. The method can generate the advertisement pictures according to the favorite style of the user, and meets the requirements of user individuation and aesthetic dynamism, thereby improving the user experience.
In some embodiments, before step S108, the method may further include: the cost function is constructed.
In some embodiments, the method may further comprise: in response to a feedback operation of a user on an advertisement picture, a value function is updated by using a Deep Deterministic Policy Gradient (DDPG). For example, after generating the advertisement picture, the advertisement picture can be displayed to the user on the front page, and the user feedback operation is obtained. The user feedback operation may include clicking or not clicking. The cost function is then updated with the user feedback according to the DDPG algorithm. The DDPG algorithm will be described in detail later with reference to the accompanying drawings.
In some embodiments, the method may further comprise: and collecting the pictures clicked by the user to update the favorite picture style of the user. For example, the pictures clicked by the user are collected and input into the deep neural network model again to generate the picture style liked by the user. The method and the device realize the collection of the favorite picture style of the user and are convenient for generating the favorite advertisement picture of the user.
FIG. 2 is a flow chart illustrating a method of generating advertisement pictures according to further embodiments of the present disclosure. As shown in FIG. 2, the method may include steps S202-S214. For example, the advertising picture of the good may include: the commodity comprises a main picture of the commodity, a decorative pattern, a layout mode, a color combination and background texture.
In step S202, a subject matting process is performed on a picture of a to-be-recommended product to obtain a subject picture of the product. The subject picture of the product is used as an input to the picture generator 220.
In step S204, a picture style corresponding to the user behavior is obtained according to the user behavior log. That is, a picture style preferred by the user is generated.
In step S206, a material space is selected according to the picture style, and material spaces required for each stage, such as material spaces for decoration patterns, layout patterns, color combinations, and background textures, are determined.
In step S208, a cost function describing the selected material is constructed, and the material that maximizes the function value of the cost function in the corresponding picture state is selected in turn from the plurality of material spaces.
For example, the cost function can be represented by Q (s, a), where s is the state (i.e., the combination of selected materials), a is the action of selecting a material from the corresponding material space (e.g., the action of selecting a material from the current material space), and Q is the cumulative value obtained by the action of selecting a material at a certain state s. Training may be performed on the cost function during reinforcement learning. For example, if a picture is simply viewed, the state immediately preceding the ending state of the picture and Q (s, a) for the last selected material are-100, and if there is click, favorite and buy activity, Q (s, a) may be equal to 20, 50 and 100, respectively.
The process of selecting material is described in detail below in conjunction with fig. 2, as follows:
first, as shown in FIG. 2, in state 1 (i.e., s)1) In this state 1, the material has not been selected yet. At this stage, the material needs to be selected from the decoration pattern. E.g. in the selection of flowers as material action a1Is time Q(s)1,a1) When the function value is maximum, the flower is selected as the decorative pattern.
Next, in state 2(s)2) In this case, state 2 includes the selected modifier pattern. At this stage, the material needs to be selected from the layout. E.g. in the act of selecting a certain pattern template as material a2Is time Q(s)2,a2) If the function value is maximum, the pattern template is selected as the layout mode.
Next, in state 3(s)3) In the case of (3), the selected decoration pattern and layout pattern are included in the state (3). At this stage, material needs to be selected from the color combination. E.g. in the act of selecting a color matching template as material a3Is time Q(s)3,a3) And if the function value is maximum, selecting the color matching template as the color combination.
Next, in state 4(s)4) In this case, the state 4 includes the selected decoration pattern, layout, and color combination. At this stage, the material needs to be selected from the background texture. E.g. in the action a of selecting a solid color of a dot as material4Is time Q(s)4,a4) When the value of the function is the maximum, the solid color of the dot is selected as the background texture.
Thus, materials of a decoration pattern, a layout mode, a color combination and a background texture which enable the value function value to be maximum are selected.
Next, an advertisement picture of the product is generated from the selected material.
In step S210, the picture generator 220 generates an advertisement picture, which is presented to the user on the front page, with user feedback (e.g., clicking or not clicking).
In step S212, the user behavior log is updated according to the user feedback, so that the pictures clicked by the user can be collected and input into the deep neural network model again to generate the favorite style of the user.
In step S214, the cost function of the picture generator 220 is updated with user feedback according to the DDPG algorithm.
For example, the following relationship may be usedNew Q(s)t,at)：
Q(St,at)←Q(St,at)+α[rt+1+γmaxaQ(st+1,a)-Q(st,at)]。 (2)
Wherein, Q(s)t,at) Indicates that the current state is stAnd the selected material action is atCost function of time, Q(s)t+1A) indicates that the next state is st+1And selects the cost function, max, when the material action is aaQ(st+1And a) represents that Q(s) is made when the material selecting action a is performedt+1A) the function value at which the function value of a) is the maximum, rt+1For immediate benefit, γ is the set attenuation coefficient and α is the learning rate. For example, in some scenarios, only the immediate benefit r is obtained during the termination statet+1Other states than 0 are all 0.
In the above relation, Q(s)t,at)+α[rt+1+γmaxaQ(st+1,a)-Q(st,at)]Updated to the latest cost function Q(s)t,at)。
To this end, methods of generating advertisement pictures according to further embodiments of the present disclosure are provided. The method can analyze the current favorite style of the user by combining the operation (such as clicking, purchasing or ignoring and the like) of the user on the advertisement picture at the client, and generate the advertisement picture which is in line with the current aesthetic value of the user, thereby meeting the requirements of the user on individuation and aesthetic dynamism and improving the user experience.
Reinforcement learning is an important branch of machine learning, and its essence is a method for continuously making automatic decision (decision making) and solving an optimal action sequence. It mainly comprises four elements: agent, state, action, and reward. The agent observes the states in the environment, selects an action based on the policy, moves to a new state and receives an instant prize, repeats the process until a final state is reached, and then obtains a cumulative prize for each state to the final state. And then repeating the process from the initial state to the final state. The policy is continually adjusted according to the jackpot over the course of the process. The strategy is repeatedly updated until an optimal strategy is reached for the agent to receive the maximum cumulative award. In the model of the present disclosure, the agent is a picture generator, the status is a currently generated pattern (i.e., a material that has been selected at the current stage), the action is to select a material in a material space corresponding to the step, the policy is a basis for status mapping action, and the reward is a user's operation on the picture (e.g., the picture is clicked for 10 points, the picture corresponds to a product purchase for 50 points, and is not clicked for-5 points, etc.). Embodiments of the present disclosure are implemented using a depth-deterministic policy gradient algorithm (DDPG). The DDPG algorithm flow may be as shown in FIG. 3.
FIG. 3 is a flow diagram illustrating a depth deterministic policy gradient algorithm according to some embodiments of the present disclosure.
DDPG is an algorithm based on the idea of an actor-critic. The actor-critic comprises two deep neural networks, an actor network and a critic network. The agent makes a decision action based on the actor network, and the environment receives this action and gives a Reward (Reward). The agent goes to a new state (state). The critic network evaluates the action based on the observed old state, new state, action and reward, which is a cost function Q. The merit function Q may reflect a cumulative award corresponding to an action in a certain state. The magnitude of the cost function can reflect whether the action is good or bad in a certain state. The cost function is calculated by a DQN (Deep Q learning) algorithm. The evaluation is returned to the actor network, which optimizes its own policy based on the evaluation. And the process is circulated until an actor network which finds the optimal strategy is trained. When the system accumulates some user feedback, the feedback is used as a strategy for rewarding and updating the picture generator on one hand, and is input into the user favorite style generation model to generate a favorite style of the user on the other hand.
Here, the DQN algorithm is an approximate implementation of the Q-learning algorithm using a deep network. The Q-learning algorithm idea is a process of repeated experience, when the experience times are close to the infinite, all action sequences and the accumulated value of the sequences can be obtained, and the action sequence with the largest accumulated value is the optimal action sequence.
FIG. 4 is a block diagram that schematically illustrates a system for generating advertising pictures, in accordance with some embodiments of the present disclosure. As shown in fig. 4, the system may include a picture processing unit 402, a genre acquisition unit 404, a material space selection unit 406, a material selection unit 408, and an advertisement picture generation unit 410.
The picture processing unit 402 may be configured to perform picture processing on a picture of a commodity to obtain a main picture of the commodity.
The style acquisition unit 404 may be configured to obtain a style of the picture corresponding to the user behavior from the user behavior log.
The material space selection unit 406 can be used to select a material space according to the picture style.
The material selection unit 408 may be configured to select, from the plurality of material spaces, a material that maximizes the function value of the cost function in the corresponding picture state. The arguments of the cost function include the picture state and the action of selecting material from the corresponding material space. The picture status represents the selected combination of material.
The advertisement picture generation unit 410 may be used to generate an advertisement picture of the goods from the selected material.
In the system of the above embodiment, the picture processing unit performs picture processing on the picture of the commodity to obtain a subject picture of the commodity. The style acquisition unit acquires a picture style corresponding to the user behavior according to the user behavior log. A material space selection unit selects a material space according to the picture style. The material selection unit sequentially selects a material that maximizes a function value of the cost function in a corresponding picture state from the plurality of material spaces. The arguments of the cost function include the picture state and the action of selecting material from the corresponding material space. The picture status represents the selected combination of material. The advertisement picture generating unit generates an advertisement picture of the commodity according to the selected material. The system can generate the advertisement pictures according to the favorite style of the user, and meets the requirements of user individuation and aesthetic dynamism, thereby improving the user experience.
In some embodiments, the material selection unit 408 may be configured to select, from the current material space, a material that maximizes the function value of the cost function in the current picture state; adding the currently selected materials into the selected material combination to obtain the next picture state; and selecting the material which enables the value function of the value function to be maximum in the next picture state from the next material space.
In some embodiments, the style obtaining unit 404 may be configured to predict a style of the picture corresponding to the user behavior from the user behavior log by using a style generation model. In the model training stage, inputting a picture with a marked picture style into a style generation model, calculating the style prediction result of the picture by the style generation model, and comparing the style prediction result with the style true result to update the parameters of the style generation model.
In some embodiments, where material is not selected in the initial stage, the state of the currently selected combination of material is a 0 vector, and the initial value of the cost function is 0.
In some embodiments, the material space may include: decorative patterns, layouts, color combinations, and background textures.
In some embodiments, the user behavior log may include pictures of merchandise that the user browses, clicks, purchases, or collects.
In some embodiments, the style acquisition unit 404 may be further configured to collect the pictures clicked by the user to update the style of the pictures liked by the user.
FIG. 5 is a block diagram that schematically illustrates a system for generating advertisement pictures, in accordance with further embodiments of the present disclosure. Similar to the system shown in fig. 4, the system shown in fig. 5 also includes a picture processing unit 402, a genre acquisition unit 404, a material space selection unit 406, a material selection unit 408, and an advertisement picture generation unit 410.
In some embodiments, as shown in FIG. 5, the system may also include a function construction unit 512. The function construction unit 512 is used for constructing the cost function.
In some embodiments, as shown in fig. 5, the system may further include an update unit 514. The updating unit 514 may be configured to update the cost function with a depth-deterministic policy gradient algorithm in response to a user feedback operation on the advertisement picture.
FIG. 6 is a block diagram that schematically illustrates a system for generating advertisement pictures, in accordance with further embodiments of the present disclosure. The system includes a memory 610 and a processor 620. Wherein:
the memory 610 may be a magnetic disk, flash memory, or any other non-volatile storage medium. The memory is used for storing instructions in the embodiments corresponding to fig. 1 and/or fig. 2.
In some embodiments, as also shown in FIG. 7, the system 700 includes a memory 710 and a processor 720. Processor 720 is coupled to memory 710 by BUS 730. The system 700 may be further coupled to an external storage device 750 via a storage interface 740 for facilitating external data transfer, and may be further coupled to a network or another computer system (not shown) via a network interface 760, which will not be described in detail herein.
In the embodiment, the data instructions are stored in the memory, and the instructions are processed by the processor, so that the advertisement pictures can be generated according to the favorite styles of the users, the requirements of individuation and aesthetic dynamism of the users are met, and the user experience is improved.
FIG. 8 is a block diagram that schematically illustrates a system for generating advertisement pictures, in accordance with further embodiments of the present disclosure. As shown in fig. 8, the system can include a client 810 and a server 820.
The client 810 may be used to present pictures of merchandise (e.g., advertising pictures) and receive user actions (e.g., user actions to browse, click, purchase, or collect). For example, the client 810 may transmit the received user behavior to the server 820 and receive the generated advertisement picture of the goods from the server 820.
The server 820 may be used to execute an operational flow for generating advertisement pictures. For example, the server 820 may execute instructions of the embodiments corresponding to fig. 1 and/or fig. 2 to generate advertisement pictures of the goods. Therefore, the advertisement pictures can be generated according to the favorite style of the user, the requirements of individuation and aesthetic dynamism of the user are met, and the user experience is improved.
In other embodiments, the present disclosure also provides a computer-readable storage medium on which computer program instructions are stored, the instructions implementing the steps of the method in the embodiment corresponding to fig. 1 and/or fig. 2 when executed by a processor. As will be appreciated by one skilled in the art, embodiments of the present disclosure may be provided as a method, apparatus, or computer program product. Accordingly, the present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present disclosure may take the form of a computer program product embodied on one or more computer-usable non-transitory storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present disclosure is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the disclosure. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Thus far, the present disclosure has been described in detail. Some details that are well known in the art have not been described in order to avoid obscuring the concepts of the present disclosure. It will be fully apparent to those skilled in the art from the foregoing description how to practice the presently disclosed embodiments.
The method and system of the present disclosure may be implemented in a number of ways. For example, the methods and systems of the present disclosure may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order for the steps of the method is for illustration only, and the steps of the method of the present disclosure are not limited to the order specifically described above unless specifically stated otherwise. Further, in some embodiments, the present disclosure may also be embodied as programs recorded in a recording medium, the programs including machine-readable instructions for implementing the methods according to the present disclosure. Thus, the present disclosure also covers a recording medium storing a program for executing the method according to the present disclosure.
Although some specific embodiments of the present disclosure have been described in detail by way of example, it should be understood by those skilled in the art that the foregoing examples are for purposes of illustration only and are not intended to limit the scope of the present disclosure. It will be appreciated by those skilled in the art that modifications may be made to the above embodiments without departing from the scope and spirit of the present disclosure. The scope of the present disclosure is defined by the appended claims.
Claims (12)
1. A method of generating an advertising picture, comprising:
carrying out picture processing on a picture of a commodity to obtain a main picture of the commodity;
obtaining a picture style corresponding to the user behavior according to the user behavior log;
selecting a material space according to the picture style;
selecting a material which enables a function value of a cost function to be maximum in a corresponding picture state from a plurality of material spaces, wherein the independent variables of the cost function comprise the picture state and an action of selecting the material from the corresponding material space, the picture state represents a selected material combination, the cost function is an accumulated value obtained by the action of selecting a certain material in a certain picture state, and the accumulated value is the sum of instant profits obtained from all the picture states in the process from the certain picture state to the ending state; and
and generating the advertisement picture of the commodity according to the selected material.
2. The method of claim 1, wherein,
the step of selecting a material from the plurality of material spaces that maximizes the function value of the cost function in the corresponding picture state includes:
selecting a material which enables the function value of the value function to be maximum under the current picture state from the current material space;
adding the currently selected materials into the selected material combination to obtain the next picture state; and
and selecting the material which enables the value function of the value function to be maximum in the next picture state from the next material space.
3. The method of claim 1, wherein prior to selecting material from the plurality of material spaces, the method further comprises: constructing the cost function.
4. The method of claim 1, wherein obtaining a picture style corresponding to user behavior from a user behavior log comprises:
predicting a picture style corresponding to the user behavior according to the user behavior log by using a style generation model;
in the model training stage, a picture with a marked picture style is input into the style generation model, the style generation model calculates a style prediction result of the picture, and the style prediction result and the style true result are compared to update parameters of the style generation model.
5. The method of claim 1, wherein,
under the condition that the materials are not selected in the initial stage, the state of the currently selected material combination is a 0 vector, and the initial value of the cost function is 0.
6. The method of claim 1, wherein,
the material space includes: decorative patterns, layouts, color combinations, and background textures.
7. The method of claim 1, wherein,
the user behavior log comprises commodity pictures browsed, clicked, purchased or collected by the user.
8. The method of claim 1, further comprising:
and updating the value function by utilizing a depth certainty strategy gradient algorithm in response to the feedback operation of the user on the advertisement picture.
9. The method of claim 1, further comprising:
and collecting the pictures clicked by the user to update the favorite picture style of the user.
10. A system for generating an advertising picture, comprising:
the picture processing unit is used for carrying out picture processing on the pictures of the commodities to obtain main pictures of the commodities;
the style acquisition unit is used for acquiring the style of the picture corresponding to the user behavior according to the user behavior log;
the material space selection unit is used for selecting a material space according to the picture style;
a material selection unit configured to select, from a plurality of material spaces, a material that maximizes a function value of a cost function in a corresponding picture state, wherein arguments of the cost function include a picture state and an action of selecting a material from the corresponding material space, the picture state representing a selected material combination, the cost function being an accumulated value obtained by performing the action of selecting a certain material at a certain picture state, the accumulated value being a sum of immediate earnings obtained for all the picture states in a process from the certain picture state to a termination state; and
and the advertisement picture generating unit is used for generating the advertisement picture of the commodity according to the selected material.
11. A system for generating an advertising picture, comprising:
a memory; and
a processor coupled to the memory, the processor configured to perform the method of any of claims 1-9 based on instructions stored in the memory.
12. A computer-readable storage medium having stored thereon computer program instructions which, when executed by a processor, implement the steps of the method of any one of claims 1 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811432961.4A CN111242654B (en) | 2018-11-28 | 2018-11-28 | Method and system for generating advertisement picture |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811432961.4A CN111242654B (en) | 2018-11-28 | 2018-11-28 | Method and system for generating advertisement picture |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111242654A CN111242654A (en) | 2020-06-05 |
| CN111242654B true CN111242654B (en) | 2021-04-30 |
Family
ID=70877373
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811432961.4A Active CN111242654B (en) | 2018-11-28 | 2018-11-28 | Method and system for generating advertisement picture |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111242654B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111833099B (en) * | 2020-06-24 | 2021-08-31 | 广州筷子信息科技有限公司 | Method and system for generating creative advertisement |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102110304A (en) * | 2011-03-29 | 2011-06-29 | 华南理工大学 | Material-engine-based automatic cartoon generating method |
| KR20120075611A (en) * | 2010-12-17 | 2012-07-09 | 주식회사 에코마케팅 | System for optimizing landing page and method thereof |
| CN103365900A (en) * | 2012-04-01 | 2013-10-23 | 阿里巴巴集团控股有限公司 | Method and device for throwing on-line material |
| CN104463779A (en) * | 2014-12-18 | 2015-03-25 | 北京奇虎科技有限公司 | Portrait caricature generating method and device |
| CN105956888A (en) * | 2016-05-31 | 2016-09-21 | 北京创意魔方广告有限公司 | Advertisement personalized display method |
| CN106445997A (en) * | 2016-07-20 | 2017-02-22 | 腾讯科技（北京）有限公司 | Information processing method and server |
| CN107330715A (en) * | 2017-05-31 | 2017-11-07 | 北京京东尚科信息技术有限公司 | The method and apparatus for selecting display advertising material |
| CN107784516A (en) * | 2016-11-29 | 2018-03-09 | 上海壹账通金融科技有限公司 | Advertisement placement method and device |
| CN108694602A (en) * | 2017-04-11 | 2018-10-23 | 阿里巴巴集团控股有限公司 | Promotional literature generation method and device |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101520782A (en) * | 2008-02-26 | 2009-09-02 | 陶鹏 | Method and system for directionally releasing special-subject information relevant to online images |
| JP5881929B2 (en) * | 2009-04-10 | 2016-03-09 | ソニー株式会社 | Server apparatus, advertisement information generation method and program |
| CN104408642B (en) * | 2014-10-29 | 2017-09-12 | 云南大学 | A kind of method for making advertising based on user experience quality |
| CN104574005B (en) * | 2015-02-15 | 2018-03-16 | 蔡耿新 | Collect augmented reality, body-sensing, the advertising display management system and method for scratching green technology |
| CN105701217B (en) * | 2016-01-13 | 2020-08-11 | 腾讯科技（深圳）有限公司 | Information processing method and server |
-
2018
- 2018-11-28 CN CN201811432961.4A patent/CN111242654B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20120075611A (en) * | 2010-12-17 | 2012-07-09 | 주식회사 에코마케팅 | System for optimizing landing page and method thereof |
| CN102110304A (en) * | 2011-03-29 | 2011-06-29 | 华南理工大学 | Material-engine-based automatic cartoon generating method |
| CN103365900A (en) * | 2012-04-01 | 2013-10-23 | 阿里巴巴集团控股有限公司 | Method and device for throwing on-line material |
| CN104463779A (en) * | 2014-12-18 | 2015-03-25 | 北京奇虎科技有限公司 | Portrait caricature generating method and device |
| CN105956888A (en) * | 2016-05-31 | 2016-09-21 | 北京创意魔方广告有限公司 | Advertisement personalized display method |
| CN106445997A (en) * | 2016-07-20 | 2017-02-22 | 腾讯科技（北京）有限公司 | Information processing method and server |
| CN107784516A (en) * | 2016-11-29 | 2018-03-09 | 上海壹账通金融科技有限公司 | Advertisement placement method and device |
| CN108694602A (en) * | 2017-04-11 | 2018-10-23 | 阿里巴巴集团控股有限公司 | Promotional literature generation method and device |
| CN107330715A (en) * | 2017-05-31 | 2017-11-07 | 北京京东尚科信息技术有限公司 | The method and apparatus for selecting display advertising material |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111242654A (en) | 2020-06-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107515909B (en) | Video recommendation method and system | |
| CN110245299B (en) | Sequence recommendation method and system based on dynamic interaction attention mechanism | |
| US10409821B2 (en) | Search result ranking using machine learning | |
| CN103502899B (en) | Dynamic prediction Modeling Platform | |
| CN107705183A (en) | Recommendation method, apparatus, storage medium and the server of a kind of commodity | |
| CN109711860A (en) | Prediction technique and device, storage medium, the computer equipment of user behavior | |
| CN108205766A (en) | Information-pushing method, apparatus and system | |
| CN110428298A (en) | A kind of shop recommended method, device and equipment | |
| CN110851699A (en) | Deep reinforcement learning-based information flow recommendation method, device, equipment and medium | |
| CN108205775A (en) | The recommendation method, apparatus and client of a kind of business object | |
| CN109102127A (en) | Method of Commodity Recommendation and device | |
| CN103646341B (en) | A kind of website provides the recommendation method and apparatus of object | |
| US20200074498A1 (en) | Systems and methods for improving social media advertising efficiency | |
| CN109582876A (en) | Tourism industry user portrait building method, device and computer equipment | |
| CN109087162A (en) | Data processing method, system, medium and calculating equipment | |
| CN109242654A (en) | A kind of item recommendation method and system | |
| CN108153792A (en) | A kind of data processing method and relevant apparatus | |
| CN110782259A (en) | Member level management method and device | |
| CN111338724A (en) | Page configuration method and device, electronic equipment and computer readable medium | |
| CN107015659A (en) | A kind of virtual try-in method of wrist-watch and system | |
| CN110602532A (en) | Entity article recommendation method, device, server and storage medium | |
| JP2008287550A (en) | Recommendation device in consideration of order of purchase, recommendation method, recommendation program and recording medium with the program recorded thereon | |
| CN113538101A (en) | Method, system and storage medium for customizing home decoration based on artificial intelligence | |
| Luo et al. | Optimal product positioning with consideration of negative utility effect on consumer choice rule | |
| CN112365283A (en) | Coupon issuing method, device, terminal equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |