Marketing method based on dynamic image generation optimization

Abstract

The present invention provides a marketing method based on dynamic image generation optimization, and relates to the technical field of marketing methods. The present invention calculates the purchasing power, shopping density and price criteria of each user according to historical shopping data, and forms a user shopping feature data set with customer shopping preference tags, calculates the switching coefficient and average browsing time of the browsing interface according to the user's historical browsing data, and forms a demand judgment feature set with the browsed commodity type and the retrieved commodity type, counts the viewing data of historical dynamic images, calculates the attractiveness value of the dynamic images, and screens the parameter features of the dynamic images to form a dynamic image feature set, optimizes the dynamic images according to the dynamic image feature set by establishing a dynamic image optimization model, calculates the matching value of the current data according to the user's shopping sign feature set, judges whether there is a shopping sign, and calculates the dynamic image push time and time interval according to the matching value.

Description

Technical Field

[0001] The present invention relates to the technical field of marketing methods, and in particular to a marketing method based on dynamic image generation optimization. Background Art

[0002] With the rapid development of the e-commerce industry, the analysis of user purchase behavior and precision marketing have become the key to improving sales efficiency and user experience. Most of the existing marketing methods rely on static advertising display or simple recommendation algorithms based on historical behavior data. Although these methods can improve conversion rates to a certain extent, they still face the following technical challenges: First, the recommendation system based on static data often cannot adapt to the user's immediate demand changes in real time, resulting in a low match between the recommended content and the user's current interests; second, the existing advertising push methods usually ignore the dynamic nature of the advertising content itself, especially the use of dynamic images, and fail to fully utilize its strong appeal and interactivity to improve user engagement; finally, although some platforms have begun to use users' browsing and purchase history data for personalized recommendations, the in-depth mining and intelligent application of these data are still in a relatively preliminary stage, and cannot truly achieve personalized, timely and precise advertising push.

[0003] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0004] The purpose of the present invention is to provide a marketing method based on dynamic image generation optimization to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A marketing method based on dynamic image generation optimization, the specific steps include:

[0007] Step 1: Obtain the user's identity information on the shopping platform to be marketed, obtain the user's historical shopping data within one year based on the identity information, and calculate the user's shopping characteristics each time he purchases goods based on the historical shopping data, including purchasing power, shopping density, and price standards;

[0008] Step 2: Obtain the user's historical browsing data and historical search data on the shopping platform based on the user's identity information, calculate the switching coefficient and average browsing time of the browsing interface each time a product is purchased based on the historical browsing data, and combine the switching coefficient and average browsing time with the search data each time a product is purchased to form a demand judgment feature;

[0009] Step 3: Count the viewing data of each user on historical animated pictures, calculate the attractiveness value of the animated pictures by viewing the data, select historical animated pictures with high attractiveness value, and filter out abnormal parameter features by calculating the normal distribution of each parameter feature of historical animated pictures with high attractiveness value. Then, match the parameter features of the filtered animated pictures with the shopping features and demand judgment features of each purchase of goods one by one.

[0010] Step 4: By building a dynamic image optimization model, according to the user's identity information, the shopping features and demand judgment features of each purchase of goods are used as the input set in the order of purchase according to the user's purchase order, and the corresponding parameter features are used as the input of the optimization model to train the model. The shopping features and demand judgment features at the current moment are used as the input of the trained model to obtain the parameter features of the dynamic image at the current moment, and the dynamic image is optimized;

[0011] Step 5: Count the user behavior characteristics, platform characteristics, and environmental characteristics of the day before the historical shopping to form a shopping sign feature set. By calculating the matching value between the current data and the historical data, determine whether there are shopping signs, and calculate the animation push time and time interval based on the matching value.

[0012] Furthermore, the historical shopping data includes the name of each purchased product, the price, quantity, time, total number of purchases, and the platform average price of the purchased product name;

[0013] The user's shopping characteristics include purchasing power, shopping intensity and price criteria;

[0014] The calculation formula of the user's purchasing power is: in, For the purchasing power of users, For the The name and price of the product purchased. For the The number of times the name of the product was purchased, is the total number of shopping items. , are all positive integers;

[0015] The calculation formula of the shopping density is: in, For shopping density, For the The time of the first purchase of the product name;

[0016] The calculation formula of the price line is: in, is the price guideline, For the The average price of product name purchased on the platform.

[0017] Furthermore, the historical browsing data includes the product type of the browsing interface, the browsing duration, the number of browsing times, and the browsing time of each browsing interface;

[0018] The search data includes the searched commodity type and the search time;

[0019] The calculation formula for the average browsing time of the interface is: in, For browsing time, is the number of views, For the View duration;

[0020] The calculation formula of the switching coefficient of the browsing interface is: in, is the switching coefficient.

[0021] Further, the viewing data includes click-through rate, number of pauses, viewing time and number of interactions;

[0022] The parameter features include the duration of the animated image, the frame change rate of the animated image, the richness of the content, and the color depth value;

[0023] The calculation formula for the attraction value of the calculation animation is: in, is the attraction value of the animated image, is the click-through rate of the animated image, is the number of pauses, is the number of interactions, To view the duration of the animation, To view the value as a basis.

[0024] Furthermore, the steps for calculating the normal distribution range of the features in the parameter features of the calculation animation are:

[0025] Calculate the parameter characteristics of historical animations, including animation duration, animation frame change rate, content richness, color depth value average and standard deviation. The specific calculation formula is: in, The parameter features of the animation are Features, The parameter features of the animation are The mean of the features, The parameter features of the animation are The standard deviation of the feature, The time is the duration of the animation, the frame rate of change, the content richness, and the color depth value. For each parameter feature data, is the total amount of data for each parameter feature, is a positive integer;

[0026] A normal distribution model of the parameter characteristics of the historical animation is established, and the calculation formula is: in, is the probability density function of each parameter feature in the parameter features of the historical animation, is the value of each parameter feature that appears in the history animation;

[0027] According to the normal distribution model of the parameter characteristics of the historical animation, the normal distribution of the parameter characteristics of the historical animation is analyzed, and the distribution probability of the parameter characteristics that need to be screened out is determined. , calculate the threshold for each parameter feature screening: in, is the threshold of each parameter feature that needs to be retained, Distribute probabilities for each parameter feature that needs to be retained.

[0028] Furthermore, the animation optimization model includes an input gate, a forget gate and an output gate based on a long short-term memory neural network;

[0029] The equation of the forget gate is: in, is the current switch state of the forget gate, represents the sigmoid function, is the forget gate weight, is the forget gate bias parameter, is the hidden state at the previous moment, Optimize the factor set for the current moment of the animation;

[0030] Determine how much of the unit state at the previous moment needs to be retained to the current moment;

[0031] The equation for the input gate is: in, is the current switch state of the input gate, is the input gate weight, is the input gate bias parameter;

[0032] in, Candidate cell states, To represent the hyperbolic tangent function, is the candidate cell weight, is the candidate cell state bias parameter;

[0033] Determine how much of the network's input data needs to be saved to the unit state at the current moment;

[0034] The update equation is: in, Current cell state, is the cell state at the previous moment;

[0035] The output gate equation is: in, is the current switch state of the output gate, is the output gate weight, is the output gate bias parameter, is the feature set of the animated image at the current moment;

[0036] Controls how much of the current unit state needs to be output to the current output value;

[0037] Among them, the mathematical expressions of the sigmoid function and the hyperbolic tangent function are: in, is the sigmoid function, is the input of the sigmoid function, is the hyperbolic tangent function.

[0038] Furthermore, the user behavior characteristics include the category of goods browsed, the number of goods, the browsing time, the type of goods retrieved, the number of retrievals, and the retrieval time;

[0039] The platform features include promotions, discounts, limited-time offers and new products;

[0040] The environmental characteristics include holidays, temperature changes, seasonal changes, and location changes;

[0041] The calculation formula of the user's behavior feature matching value is: in, is the ratio of each feature of user behavior characteristics to the overlap of historical user behavior characteristics, is the number of features of user behavior characteristics, For the historical data The number of user behavior features with the same number of features before the purchase, is the behavior feature matching value, The ratio of overlapping user behavior features is used to determine the threshold.

[0042] Furthermore, the calculation formula of the platform feature matching value is: in, is the ratio of each feature of the platform characteristics to the historical platform characteristics, is the number of features of the platform characteristics, For the historical data The number of platform features that are the same as the number of features before the purchase, is the platform feature matching value, It is the ratio judgment threshold of platform feature overlap.

[0043] Furthermore, the calculation formula of the environmental feature matching value is: in, is the ratio of each feature of the environmental characteristics to the historical environmental characteristics, is the number of features of the environmental characteristics, For the historical data The number of environmental features that are the same as the number of features before the purchase, is the environmental feature matching value, It is the ratio judgment threshold of the environmental feature overlap.

[0044] Furthermore, the calculation formula of the matching value is: in, is the matching value, are the weights of the matching values ​​of user behavior characteristics, platform characteristics, and environmental characteristics, respectively. ;

[0045] when When the user is detected as having shopping characteristics, the animated picture will be pushed;

[0046] in, is the judgment threshold of the matching value;

[0047] The calculation formula for the push time interval is: in, is the push time interval, is the time interval adjustment constant, Adjust the bias constant for time.

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows: the purchasing power, shopping density and price guideline of each user are calculated according to the historical shopping data, and a user shopping feature data set is formed with the user shopping preference tag; the switching coefficient of the browsing interface and the average browsing time are calculated according to the user's historical browsing data, and a demand judgment feature set is formed with the browsed commodity type and the retrieved commodity type; the viewing data of the historical animated images are counted, the attractiveness value of the animated images is calculated, and the parameter features of the animated images are screened to form an animated image feature set; by establishing an animated image optimization model, the animated images are optimized according to the animated image feature set; according to the user's shopping sign feature set, the matching value of the current data is calculated to determine whether there is a shopping sign, and the animated image push time and time interval are calculated according to the matching value;

[0049] The present invention provides a new method for optimizing animated image advertisements based on a multi-level data analysis and optimization model, and based on multiple dimensions such as user identity information, historical shopping behavior, browsing data, and dynamic image viewing data. The normal distribution is used to ensure a reasonable range of features, and to filter out extreme or uncommon data that may be detrimental to marketing results, thereby helping to optimize the overall performance of animated images. By establishing an animated image optimization model, combined with the user's shopping behavior characteristics, demand judgment characteristics, and the attractiveness of historical animated images, the display time and interval of advertisements are accurately predicted and optimized, greatly improving the relevance and timeliness of advertisements. This method solves the problem that traditional advertising push systems cannot be accurately optimized based on the user's immediate needs;

[0050] The present invention can also more accurately capture the user's shopping intentions and preferences by accurately analyzing the user's historical shopping data, browsing behavior data and animated image attraction value, and improve the personalization and relevance of advertising content. At the same time, by intelligently calculating the push time and time interval, it can push ads at the moment when the user is most likely to buy, reducing unnecessary advertising interruptions and improving the conversion rate of ads. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a schematic diagram of the overall method flow of the present invention. DETAILED DESCRIPTION

[0052] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments.

[0053] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0054] Example

[0055] See also Figure 1 , the present invention provides a technical solution:

[0056] A marketing method based on dynamic image generation optimization, the specific steps include:

[0057] Step 1: Obtain the user's identity information on the shopping platform to be marketed, obtain the user's historical shopping data within one year based on the identity information, and calculate the user's shopping characteristics each time he purchases goods based on the historical shopping data, including purchasing power, shopping intensity, and price guidelines.

[0058] The user shopping feature dataset can accurately reflect each user's consumption habits, purchasing power and preferences by analyzing information such as user purchasing power, shopping frequency, price preferences, etc. This information is crucial for generating personalized marketing strategies when using dynamic images for influence.

[0059] Through the user's shopping feature data, we can better understand the user's needs, aesthetics and preferences. Based on this data, dynamic image generation technology can generate personalized advertisements, product display pictures or promotional images in real time. These images can cater to the user's personal preferences in terms of appearance style, function, color matching, etc. By analyzing each user's shopping history data, the user's shopping time cycle and the category of high-frequency purchases can be identified. For example, if a user suddenly becomes interested in home products, the system can automatically generate home-related advertising images based on their recent search and purchase behaviors and display them on the user's device.

[0060] Shopping feature data plays a vital role in the marketing method of dynamic image generation optimization. It provides basic data support for dynamic image generation technology through accurate user portraits and in-depth analysis of user behavior, so that marketing content can achieve better results in personalization, precision, and dynamism. Specific benefits include personalized advertising generation, real-time adaptation to user needs, multi-dimensional recommendations, increased participation, and enhanced visual experience, which will greatly promote the improvement of marketing effectiveness, thereby increasing user conversion rates and brand loyalty.

[0061] In this embodiment, the historical shopping data includes the name of each purchased product, the price, quantity, time, total number of purchases, and the platform average price of the purchased product name;

[0062] The user's shopping characteristics include purchasing power, shopping intensity and price criteria;

[0063] The calculation formula of the user's purchasing power is: in, For the purchasing power of users, For the The name and price of the product purchased. For the The number of times the name of the product was purchased, is the total number of shopping items. , are all positive integers;

[0064] The calculation formula of the shopping density is: in, For shopping density, For the The time of the first purchase of the product name;

[0065] The calculation formula of the price line is: in, is the price guideline, For the The average price of product name purchased on the platform.

[0066] Price criteria refer to the user's sensitivity and acceptance range for product prices. Based on the user's price criteria, the system can generate advertising content that meets their price preferences. Users with lower price criteria will see moderately priced products, while users with higher price criteria will see high-priced products or high-end brands. Dynamic images can automatically adjust the display of product prices based on the user's price criteria, making the advertising content more in line with the user's purchasing potential.

[0067] Step 2: Obtain the user's historical browsing data and historical search data on the shopping platform based on the user's identity information, calculate the switching coefficient and average browsing time of the browsing interface each time a product is purchased based on the historical browsing data, and combine the switching coefficient and average browsing time with the search data for each time a product is purchased to form a demand judgment feature.

[0068] In this embodiment, the historical browsing data includes the product type of the browsing interface, the browsing duration, the number of browsing times, and the browsing time of each browsing interface;

[0069] The search data includes the searched commodity type and the search time;

[0070] The calculation formula for the average browsing time of the interface is: in, For browsing time, is the number of views, For the View duration.

[0071] Average browsing time refers to the average time users stay on each product page. This indicator reflects the depth of user interest in the product. If users stay on a product page for a long time, it usually means that they have a strong interest in the product and may be making a purchase decision. If users browse each product page for a short time, it means that they are less focused or are browsing multiple products quickly. In this case, the dynamic duration of the animated image should be kept short to quickly convey basic product information, avoid slowing down the user's browsing speed, and encourage them to continue browsing as soon as possible. Too long animated image playback may make users bored or interrupt their browsing rhythm.

[0072] The calculation formula of the switching coefficient of the browsing interface is: in, is the switching coefficient.

[0073] The switching coefficient of the browsing interface refers to the frequency with which users switch between different products or pages when browsing products. This indicator reveals the user's sensitivity and interest in changes in product categories, page content, etc. If users frequently switch pages or products, it means that their attention may be high, but they may not necessarily stay on a certain product for too long. When users frequently switch browsing interfaces, it means that their interests may be scattered and they may not have a deep interest in a specific product. In this case, the dynamic duration of the animated image should be short. Short-term animated image playback can quickly convey product information without slowing down the user's browsing rhythm. If the animated image is played for too long, it may interfere with the user's browsing and cause the user to quickly skip the ad.

[0074] The function of the demand judgment feature set is to infer the user's current needs and potential purchase intentions by analyzing the user's browsing behavior, time and product type. It reveals the user's interaction pattern on the platform and provides the system with a data basis for dynamically adjusting marketing strategies.

[0075] By analyzing the types of browsed goods and retrieved goods, the system can accurately capture the user's current needs. For example, when a user frequently browses a certain type of goods such as sports shoes and electronic products, the system can prioritize these goods in dynamic image ads and dynamically adjust the ad content to ensure that it meets the user's actual needs. Browsing time and interface switching coefficient can reflect the user's attention to and purchase intention for a certain product. If the user stays for a long time and switches interfaces many times, it indicates that the user may be making a purchase decision. At this time, the system can push promotional information, coupons or inventory reminders about the product through dynamic image ads to further stimulate the user's desire to buy.

[0076] Step 3: Count the viewing data of each user on historical animated pictures, calculate the attractiveness value of the animated pictures by viewing the data, select historical animated pictures with high attractiveness value, and filter out abnormal parameter features by calculating the normal distribution of each parameter feature of historical animated pictures with high attractiveness value. Then, make a one-to-one correspondence between the parameter features of the filtered animated pictures and the shopping features and demand judgment features of each purchase of goods.

[0077] In this embodiment, the viewing data includes click rate, number of pauses, viewing time and number of interactions;

[0078] The parameter features include the duration of the animated image, the frame change rate of the animated image, the richness of the content, and the color depth value;

[0079] The calculation formula for the attraction value of the calculation animation is: in, is the attraction value of the animated image, is the click-through rate of the animated image, is the number of pauses, is the number of interactions, To view the duration of the animation, To view the value as a basis.

[0080] The attraction value refers to a comprehensive score or index that measures the ability of a GIF to attract user attention and generate interaction. This indicator is usually calculated by analyzing the interaction behavior data between users and the GIF (such as click-through rate, number of pauses, viewing time, and number of interactions). The attraction value directly reflects the effectiveness of the GIF in attracting user viewing and participation. GIFs with high attraction values ​​are usually able to arouse user interest, increase viewing time, and promote interaction and conversion.

[0081] In this embodiment, the steps for calculating the normal distribution range of the features in the parameter features of the calculation animation are:

[0082] Calculate the parameter characteristics of historical animations, including animation duration, animation frame change rate, content richness, color depth value average and standard deviation. The specific calculation formula is: in, The parameter features of the animation are Features, The parameter features of the animation are The mean of the features, The parameter features of the animation are The standard deviation of the feature, The time is the duration of the animation, the frame rate of change, the content richness, and the color depth value. For each parameter feature data, is the total amount of data for each parameter feature, is a positive integer;

[0083] A normal distribution model of the parameter characteristics of the historical animation is established, and the calculation formula is: in, is the probability density function of each parameter feature in the parameter features of the historical animation, is the value of each parameter feature that appears in the history animation;

[0084] According to the normal distribution model of the parameter characteristics of the historical animation, the normal distribution of the parameter characteristics of the historical animation is analyzed, and the distribution probability of the parameter characteristics that need to be screened out is determined. , calculate the threshold for each parameter feature screening: in, is the threshold of each parameter feature that needs to be retained, Distribute probabilities for each parameter feature that needs to be retained.

[0085] Normal distribution is the most commonly used statistical distribution form in data analysis. In many natural phenomena and human behaviors, data often presents the characteristics of normal distribution, that is, most data points are concentrated near the mean value, and there are relatively few extreme high and low values. By filtering with normal distribution, we can find the typical and most common parameter features of animated images, thereby avoiding the use of those animated images that are too extreme and do not conform to the viewing habits of most users. If the parameter features of the animated image are not screened, the final animated image effect and the user's viewing experience may be affected by extreme data (such as too long animation duration, too high frame change rate, etc.). By filtering with normal distribution, we can ensure that the various features of the animated image are within a reasonable range, thereby reducing the deviation caused by extreme data and avoiding the design of animated images that users do not like.

[0086] Screening the parameter features of animated images through normal distribution is a data analysis and optimization method. Its purpose is to effectively evaluate the different features of animated images, such as animation duration, frame rate, content richness, and color depth value, and screen out the most attractive and effective animated features. The key idea behind this method is to use normal distribution to ensure a reasonable range of features and filter out extreme or uncommon data that may be detrimental to marketing results, thereby helping to optimize the overall performance of animated images.

[0087] The animated image feature set is a key data set formed by screening and processing the viewing data and parameter features of historical animated images. It represents the various performance characteristics of animated images, and through the screening method of normal distribution, it further helps marketers understand which features of animated images are most attractive to users. By using the animated image feature set, brands can optimize advertising strategies, improve advertising accuracy, user interaction rate, and ultimately conversion rate, thereby achieving more effective dynamic image generation optimization marketing methods.

[0088] Step 4: By establishing a dynamic image optimization model, according to the user's identity information, the shopping characteristics and demand judgment characteristics of each purchase are used as the input set in the order of purchase according to the user's purchase order, and the corresponding parameter characteristics are used as the input of the optimization model to train the model. The shopping characteristics and demand judgment characteristics at the current moment are used as the input of the trained model to obtain the parameter characteristics of the dynamic image at the current moment, and the dynamic image is optimized.

[0089] By building a dynamic image optimization model through a long short-term memory neural network, the effect of dynamic images in marketing can be significantly improved. When processing time series data, the long short-term memory neural network can effectively capture the temporal changes in user behavior, which is crucial for the personalized optimization of dynamic images. For example, e-commerce platforms can use long short-term memory neural networks to model users' past browsing, clicking, purchasing and other behaviors, thereby predicting changes in users' interests at different time points. Based on these predictions, the system can optimize the dynamic image content displayed to users in real time to make it more in line with the user's current needs and preferences. For example, if a user has frequently browsed electronic products in the past week, the long short-term memory neural network can predict that the user is more interested in the newly released electronic product dynamic images, thereby pushing the corresponding optimized dynamic images to improve user engagement and purchase conversion rates.

[0090] This method can not only adjust the recommended content according to the user's immediate interests, but also continuously optimize the recommendation effect in long-term use. By combining historical behavior and real-time interaction data, the long short-term memory neural network can continuously adjust the optimization strategy, thereby improving the user's viewing experience and the user stickiness of the platform.

[0091] Optimizing animated images through long short-term memory neural networks can not only improve user experience, but also increase the effectiveness of marketing activities. For example, on e-commerce platforms, through in-depth analysis of behavioral data of different user groups, long short-term memory neural networks can generate customized animated image ads that best meet the shopping needs of each user. This precise dynamic adjustment capability helps merchants better reach potential customers and push the right content at the right time.

[0092] In this embodiment, the animation optimization model is based on a long short-term memory neural network and includes an input gate, a forget gate, and an output gate;

[0093] The equation of the forget gate is: in, is the current switch state of the forget gate, represents the sigmoid function, is the forget gate weight, is the forget gate bias parameter, is the hidden state at the previous moment, Optimize the factor set for the current moment of the animation;

[0094] Determine how much of the unit state at the previous moment needs to be retained to the current moment;

[0095] The equation for the input gate is: in, is the current switch state of the input gate, is the input gate weight, is the input gate bias parameter;

[0096] in, Candidate cell states, To represent the hyperbolic tangent function, is the candidate cell weight, is the candidate cell state bias parameter;

[0097] Determine how much of the network's input data needs to be saved to the unit state at the current moment;

[0098] The update equation is: in, Current cell state, is the cell state at the previous moment;

[0099] The output gate equation is: in, is the current switch state of the output gate, is the output gate weight, is the output gate bias parameter, is the feature set of the animated image at the current moment;

[0100] Controls how much of the current unit state needs to be output to the current output value;

[0101] Among them, the mathematical expressions of the sigmoid function and the hyperbolic tangent function are: in, is the sigmoid function, is the input of the sigmoid function, is the hyperbolic tangent function.

[0102] Step 5: Count the user behavior characteristics, platform characteristics, and environmental characteristics of the day before the historical shopping to form a shopping sign feature set. By calculating the matching value between the current data and the historical data, determine whether there are shopping signs, and calculate the animation push time and time interval based on the matching value.

[0103] In this embodiment, the user behavior characteristics include the category of browsed goods, the number of goods, the browsing time, the type of searched goods, the number of searches, and the search time;

[0104] The platform features include promotions, discounts, limited-time offers and new products;

[0105] The environmental characteristics include holidays, temperature changes, seasonal changes, and location changes;

[0106] The calculation formula of the user's behavior feature matching value is: in, is the ratio of each feature of user behavior characteristics to the overlap of historical user behavior characteristics, is the number of features of user behavior characteristics, For the historical data The number of user behavior features with the same number of features before the purchase, is the behavior feature matching value, The ratio of overlapping user behavior features is used to determine the threshold.

[0107] The user behavior feature matching value represents the matching degree between the current user's behavior features (such as browsed product categories, product quantity, browsing time, retrieved product type, retrieval times, retrieval time, etc.) and similar behaviors in historical data. By comparing the similarity between the user's current behavior and similar shopping behavior patterns in the past, it can be inferred whether the user is showing a shopping tendency.

[0108] In this embodiment, the calculation formula of the platform feature matching value is: in, is the ratio of each feature of the platform characteristics to the historical platform characteristics, is the number of features of the platform characteristics, For the historical data The number of platform features that are the same as the number of features before the purchase, is the platform feature matching value, It is the ratio judgment threshold of platform feature overlap.

[0109] The platform feature matching value reflects the user's purchase intention in a platform-specific promotion environment. If the current platform features (such as discounts, limited-time offers, etc.) are highly similar to the user's purchase pattern in historical data, it means that the user may have a strong shopping tendency under the current platform promotion.

[0110] In this embodiment, the calculation formula of the environmental feature matching value is: in, is the ratio of each feature of the environmental characteristics to the historical environmental characteristics, is the number of features of the environmental characteristics, For the historical data The number of environmental features that are the same as the number of features before the purchase, is the environmental feature matching value, It is the ratio judgment threshold of the environmental feature overlap.

[0111] The environmental feature matching value reflects the degree of fit between the current environmental factors, such as holidays and temperature changes, and the historical shopping environment. If the environmental matching value is high, it means that the user's shopping behavior is similar to the historical shopping behavior in similar environments. At this time, pushing relevant dynamic image ads may be more effective.

[0112] In this embodiment, the calculation formula of the matching value is: in, is the matching value, are the weights of the matching values ​​of user behavior characteristics, platform characteristics, and environmental characteristics, respectively. ;

[0113] when When the user is detected as having shopping characteristics, the animated picture will be pushed;

[0114] in, is the judgment threshold of the matching value.

[0115] The match value is a key indicator to measure the similarity between current user behavior data and historical data. It reflects the degree of fit between the current user behavior characteristics, platform characteristics, environmental characteristics, etc. and the historical data pattern, and then helps the system determine whether the user has a purchase intention or shopping signs. A high match value means that the user's behavior is highly similar to the historical shopping behavior, indicating that they are showing a tendency to buy. At this time, the marketing system can push animated image ads for related products to capture the moment when the user is about to make a decision. Accurate push can improve the relevance and attractiveness of ads and reduce the interference of invalid ads. The time and time interval of push can be calculated based on the match value, and the frequency and timing of marketing activities can be accurately adjusted. For example, when the match value is high, the system may push more dynamic image ads and shorten the push interval to maximize the user's attention; when the match value is low, the push frequency will be reduced or delayed to avoid disturbing the user.

[0116] The calculation formula for the push time interval is: in, is the push time interval, is the time interval adjustment constant, Adjust the bias constant for time.

[0117] If the interval between ad pushes is too short, users may become fatigued by the ad; if the interval is too long, users may forget the ad content or miss the right time to buy. By calculating the optimal time interval from historical data, these problems can be effectively avoided, ensuring that the ad is seen by users at the best time while reducing ad repetition.

[0118] Through intelligent calculation of push time and time interval, ads can not only appear in front of users at the right time, but also minimize the waste of advertising budget. For example, through precise matching, the platform can avoid pushing ads to inactive user groups or pushing ads at irrelevant time periods, thus ensuring that every exposure of the ad can generate higher returns.

[0119] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0120] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. Those skilled in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed by hardware or software methods depends on the specific application and design constraints of the technical solution.

[0121] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0122] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A marketing method based on dynamic image generation optimization, characterized in that: The specific steps include: Step 1: Obtain the user's identity information on the shopping platform to be marketed, obtain the user's historical shopping data within one year based on the identity information, and calculate the user's shopping characteristics each time he purchases goods based on the historical shopping data, including purchasing power, shopping density, and price standards; Step 2: Obtain the user's historical browsing data and historical search data on the shopping platform based on the user's identity information, calculate the switching coefficient and average browsing time of the browsing interface each time a product is purchased based on the historical browsing data, and combine the switching coefficient and average browsing time with the search data each time a product is purchased to form a demand judgment feature; Step 3: Count the viewing data of each user on historical animated pictures, calculate the attractiveness value of the animated pictures by viewing the data, select historical animated pictures with high attractiveness value, and filter out abnormal parameter features by calculating the normal distribution of each parameter feature of historical animated pictures with high attractiveness value. Then, match the parameter features of the filtered animated pictures with the shopping features and demand judgment features of each purchase of goods one by one. Step 4: By building a dynamic image optimization model, according to the user's identity information, the shopping features and demand judgment features of each purchase of goods are used as input sets in the order of purchase according to the user's purchase order, and the corresponding parameter features are used as inputs of the optimization model to train the model. The shopping features and demand judgment features at the current moment are used as inputs of the trained model to obtain the parameter features of the dynamic image at the current moment, and the dynamic image is optimized; Step 5: Count the user behavior characteristics, platform characteristics, and environment characteristics of the day before the historical shopping to form a shopping sign feature set. By calculating the matching value between the current data and the historical data, determine whether there is a shopping sign, and calculate the time and time interval for pushing the animated image based on the matching value. The viewing data includes click-through rate, number of pauses, viewing time and number of interactions; The parameter features include the duration of the animated image, the frame change rate of the animated image, the richness of the content, and the color depth value; The calculation formula for the attraction value of the calculation animation is: in, is the attraction value of the animated image, is the click-through rate of the animated image, is the number of pauses, is the number of interactions, To view the duration of the animation, To view the value as a basis.

2. The marketing method based on dynamic image generation optimization according to claim 1, characterized in that: The historical shopping data includes the name, price, quantity, time, total number of times the product was purchased, and the platform average price of the product name purchased each time; The user's shopping characteristics include purchasing power, shopping intensity and price criteria; The calculation formula of the user's purchasing power is: in, For the purchasing power of users, For the The name and price of the product purchased. For the The number of times the name of the product was purchased, is the total number of shopping items. , All are positive integers; The calculation formula of the shopping density is: in, For shopping density, For the The time of the first purchase of the product name; The calculation formula of the price line is: in, is the price guideline, For the The average price of product name purchased on the platform.

3. The marketing method based on dynamic image generation optimization according to claim 1, characterized in that: The historical browsing data includes the product type of the browsing interface, the browsing duration, number of browsing times, and browsing time of each browsing interface; The search data includes the searched commodity type and the search time; The calculation formula for the average browsing time of the interface is: in, For browsing time, is the number of views, For the View duration; The calculation formula of the switching coefficient of the browsing interface is: in, is the switching coefficient.

4. The marketing method based on dynamic image generation optimization according to claim 1, characterized in that: The steps for calculating the normal distribution range of the features in the parameter features of the calculation animation are: Calculate the parameter characteristics of historical animations, including animation duration, animation frame change rate, content richness, color depth value average and standard deviation. The specific calculation formula is: in, The parameter features of the animation are Features, The parameter features of the animation are The mean of the features, The parameter features of the animation are The standard deviation of the feature, The time is the duration of the animation, the frame rate of change, the content richness, and the color depth value. For each parameter feature data, is the total amount of data for each parameter feature, is a positive integer; A normal distribution model of the parameter characteristics of the historical animation is established, and the calculation formula is: in, is the probability density function of each parameter feature in the parameter features of the historical animation, is the value of each parameter feature that appears in the history animation; According to the normal distribution model of the parameter characteristics of the historical animation, the normal distribution of the parameter characteristics of the historical animation is analyzed, and the distribution probability of the parameter characteristics that need to be screened out is determined. , calculate the threshold for each parameter feature screening: in, is the threshold value of each parameter feature that needs to be retained, Distribute probabilities for each parameter feature that needs to be retained.

5. The marketing method based on dynamic image generation optimization according to claim 1, characterized in that: The dynamic image optimization model is based on a long short-term memory neural network and includes an input gate, a forget gate and an output gate; The equation of the forget gate is: in, is the current switch state of the forget gate, represents the sigmoid function, is the forget gate weight, is the forget gate bias parameter, is the hidden state at the previous moment, Optimize the factor set for the current moment of the animation; Determine how much of the unit state at the previous moment needs to be retained to the current moment; The equation for the input gate is: in, is the current switch state of the input gate, is the input gate weight, is the input gate bias parameter; in, is the candidate cell state, To represent the hyperbolic tangent function, is the candidate cell weight, is the candidate cell state bias parameter; Determine how much of the network's input data needs to be saved to the unit state at the current moment; The update equation is: in, Current cell state, is the cell state at the previous moment; The output gate equation is: in, is the current switch state of the output gate, is the output gate weight, is the output gate bias parameter, is the feature set of the animated image at the current moment; Controls how much of the current unit state needs to be output to the current output value; Among them, the mathematical expressions of the sigmoid function and the hyperbolic tangent function are: in, is the sigmoid function, is the input of the sigmoid function, is the hyperbolic tangent function.

6. The marketing method based on dynamic image generation optimization according to claim 1, characterized in that: The user behavior characteristics include the category of goods browsed, the number of goods, the browsing time, the type of goods searched, the number of searches, and the search time; The platform features include promotions, discounts, limited-time offers and new products; The environmental characteristics include holidays, temperature changes, seasonal changes, and location changes; The calculation formula of the user's behavior feature matching value is: in, is the ratio of each feature of user behavior characteristics to the overlap of historical user behavior characteristics, is the number of features of user behavior characteristics, For the historical data The number of user behavior features with the same number of features before the purchase, is the behavior feature matching value, The ratio of overlapping user behavior features is used to determine the threshold.

7. The marketing method based on dynamic image generation optimization according to claim 6, characterized in that: The calculation formula for the platform feature matching value is: in, is the ratio of each feature of the platform characteristics to the historical platform characteristics, is the number of features of the platform characteristics, For the historical data The number of platform features that have the same number of features as before the purchase, is the platform feature matching value, It is the ratio judgment threshold of platform feature overlap.

8. The marketing method based on dynamic image generation optimization according to claim 7, characterized in that: The calculation formula of the environmental feature matching value is: in, is the ratio of each feature of the environmental characteristics to the historical environmental characteristics, is the number of features of the environmental characteristics, For the historical data The number of environmental features that are the same as the number of features before the purchase, is the environmental feature matching value, It is the ratio judgment threshold of the environmental feature overlap.

9. The marketing method based on dynamic image generation optimization according to claim 8, characterized in that: The calculation formula of the matching value is: in, is the matching value, are the weights of the matching values ​​of user behavior characteristics, platform characteristics, and environmental characteristics, respectively. ; when When the user is detected as having shopping characteristics, the animated picture will be pushed; in, is the judgment threshold of the matching value; The calculation formula for the push time interval is: in, is the push time interval, The constant for the time interval adjustment, Adjust the bias constant for time.

Images