An advertisement delivery strategy optimization method and system based on AI learning

Abstract

The embodiment of the application relates to the field of artificial intelligence technology, and discloses an advertisement putting strategy optimization method and system based on AI learning. The method first captures real-time state parameters of an external environment by using a multi-source environment sensor, combines big data analysis with hot online shopping data, and then quantifies the matching degree of each candidate advertisement and the environment state by using a pre-training model to generate a first putting strategy. Then, in the advertisement display process, implicit behavior data such as user gaze duration is taken as a feedback index to perform secondary optimization on the first putting strategy to obtain a second putting strategy. In this way, by combining environment parameters and online shopping hot spot data, the initial strategy is optimized by using model scoring and sorting, the advertisement putting is more suitable for the environment state and online shopping consumption trend, the adaptability of the advertisement putting to the scene environment and online shopping hot spot is greatly improved, the advertisement putting effect and user attention are improved, and intelligent and accurate optimization of the advertisement putting strategy is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to an advertisement delivery strategy optimization method and system based on AI learning. BACKGROUND

[0002] In the era of digital marketing, the precision and efficiency of advertisement delivery directly affect the operational efficiency of commercial entities. Traditional mall advertisement delivery strategies usually arrange content based on fixed time periods, preset rules, or manual experience, which cannot dynamically adapt to environmental changes and real-time fluctuations in user behavior. For example, seasonal changes, weather changes, or unexpected events can cause significant changes in consumer shopping preferences, and static advertisement strategies cannot capture such environmental correlations in a timely manner, resulting in resource waste and low conversion rates. At the same time, the rapid development of online consumer behavior has led to the integration of online and offline shopping decisions, but the offline advertisement system lacks the ability to integrate real-time online shopping trends, resulting in insufficient matching of advertisement content and user potential needs. SUMMARY

[0003] The main purpose of the present application is to provide an advertisement delivery strategy optimization method and system based on AI learning, which aims to solve the technical problem of insufficient matching of advertisement content and user potential needs caused by static advertisement delivery in the prior art.

[0004] To achieve the above-mentioned purpose, in a first aspect, an advertisement delivery strategy optimization method based on AI learning is provided in the embodiments of the present application, which is applied to an advertisement delivery strategy optimization system including an environment sensor and an advertisement delivery screen. The method comprises:

[0005] Obtaining environment state parameters collected by the environment sensor in real time, the environment state parameters including at least one of indoor environment parameters and outdoor environment parameters;

[0006] Obtaining hot online shopping data matched with the environment state parameters according to the environment state parameters;

[0007] Inputting the environment state parameters and the hot online shopping data into a pre-trained advertisement matching model to output matching score of each candidate advertisement;

[0008] Prioritizing each candidate advertisement according to the matching score to generate a first advertisement delivery strategy to preliminarily optimize the initial advertisement delivery strategy;

[0009] Controlling the playing content of the advertisement delivery screen according to the first advertisement delivery strategy, and collecting user gaze duration data in real time during the playing process;

[0010] According to the user gaze duration data, the first advertisement delivery strategy is re-optimized to obtain a second advertisement delivery strategy.

[0011] In a possible implementation, the environment sensor includes an indoor environment sensor and an outdoor environment sensor, and the environment state parameter collected in real time by the environment sensor includes:

[0012] The indoor environment state parameter collected in real time by the indoor environment sensor includes at least an indoor people flow density;

[0013] The outdoor environment state parameter collected in real time by the outdoor environment sensor includes at least one of an environment temperature and an environment rainfall.

[0014] In a possible implementation, the hot online shopping data matched with the environment state parameter is obtained according to the environment state parameter, and the hot online shopping data includes:

[0015] When the indoor people flow density is greater than or equal to a people flow density threshold value and the environment temperature is greater than or equal to a temperature threshold value, hot-selling commodity data of online shopping in high-temperature weather is obtained by using a reinforcement learning search algorithm;

[0016] Or, when the indoor people flow density is greater than or equal to a people flow density threshold value and the environment rainfall is greater than or equal to a rainfall threshold value, hot-selling commodity data of online shopping in rainy weather is obtained by using a reinforcement learning search algorithm.

[0017] In a possible implementation, the environment state parameter and the hot online shopping data are input into a pre-trained advertisement matching model, and a matching degree score of each candidate advertisement is output, and the method includes:

[0018] The environment state parameter is input into an environment subnetwork of the pre-trained advertisement matching model to obtain an environment attribute score of each candidate advertisement;

[0019] The hot online shopping data is input into a hot subnetwork of the pre-trained advertisement matching model to obtain a hot attribute score of each candidate advertisement;

[0020] The environment attribute score and the hot attribute score are weighted and summed to obtain the matching degree score of each candidate advertisement.

[0021] In a possible implementation, before the environment attribute score and the hot attribute score are weighted and summed to obtain the matching degree score of each candidate advertisement, the method further includes:

[0022] A change rate of the environment state parameter and a trend acceleration of the hot online shopping data are obtained;

[0023] The rate of change of the environment state parameter and the trend acceleration of the hot spot online shopping data are input into a dynamic selection network to generate a dynamic weight coefficient, wherein the dynamic selection network satisfies the following expression:

[0024] ;

[0025] ;

[0026] ;

[0027] In the formula, is a dynamic weight of the environment attribute score, is a dynamic weight of the hot spot attribute score; is the rate of change of the environment state parameter, is a reference rate of change of the environment state parameter, is a reference weight of the environment attribute score; is the trend acceleration of the hot spot online shopping data, is a reference trend acceleration of the hot spot online shopping data, is a reference weight of the hot spot attribute score; is an alarm value of the rate of change of the environment state parameter, is an alarm value of the trend acceleration of the hot spot online shopping data.

[0028] In a possible implementation, the candidate advertisements are prioritized according to the matching degree scores, and a first advertisement delivery strategy is generated to preliminarily optimize the initial advertisement delivery strategy, including:

[0029] The candidate advertisements are prioritized according to the matching degree scores to obtain an advertisement delivery playlist;

[0030] The display time length of each candidate advertisement in the advertisement delivery playlist is determined according to the matching degree score of each candidate advertisement.

[0031] In a possible implementation, the display time length of each candidate advertisement in the advertisement delivery playlist is determined according to the matching degree score of each candidate advertisement, including:

[0032] The reference display time length of each candidate advertisement is determined according to the total time length constraint of the advertisement and the number of candidate advertisements;

[0033] The actual display time length of each candidate advertisement is obtained by dynamically adjusting the reference display time length according to the matching degree score of each candidate advertisement.

[0034] In a possible implementation, the first advertisement delivery strategy is further optimized according to the user gaze time length data to obtain a second advertisement delivery strategy, including:

[0035] determine a delivery efficiency value of each candidate advertisement according to the user gaze duration data and an actual display duration of each candidate advertisement, wherein the delivery efficiency value is a ratio of a user gaze duration average value to the actual display duration;

[0036] re-optimize the first advertisement delivery strategy according to the delivery efficiency value of each candidate advertisement to obtain a second advertisement delivery strategy.

[0037] In a possible implementation, re-optimizing the first advertisement delivery strategy according to the delivery efficiency value of each candidate advertisement to obtain a second advertisement delivery strategy includes:

[0038] when the delivery efficiency value of a candidate advertisement exceeds 1.2 times of a mean value of similar advertisements in continuous N monitoring periods, triggering a gradual display duration gain strategy;

[0039] when the delivery efficiency value of a candidate advertisement is lower than 0.8 times of the mean value of similar advertisements in continuous M monitoring periods, triggering a display duration hierarchical attenuation strategy, wherein 5 > N > M > 3.

[0040] In a second aspect, the embodiments of the present application further provide an advertisement delivery strategy optimization system, including a memory and a processor, the memory is used to store program code, and the processor is used to call the program code to execute the method in the first aspect.

[0041] Compared with the prior art, the advertisement delivery strategy optimization method based on AI learning provided in the embodiments of the present application first captures real-time state parameters of an external environment by using a multi-source environmental sensor, combines big data analysis with hot online shopping data, and then quantifies the matching degree of each candidate advertisement and the environmental state by using a pre-trained model to generate a first delivery strategy; subsequently, in the process of advertisement display, the first delivery strategy is optimized twice to obtain a second delivery strategy by taking user gaze duration and other implicit behavior data as feedback indicators. In this way, by combining environmental parameters and online shopping hot spot data, the initial strategy is optimized by using model scoring and ranking, so that the advertisement delivery is more suitable for the environmental state and the online shopping consumption trend, greatly improving the adaptability of the advertisement delivery to the scene environment and the online shopping hot spot; and the user interest can be accurately captured, the attractiveness and pertinence of the advertisement are enhanced, and thus the advertisement delivery effect and the user attention are improved, realizing intelligent and accurate optimization of the advertisement delivery strategy. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only represent some of the embodiments of the present application, and all other drawings obtained by those of ordinary skill in the art without creative effort based on these drawings also belong to the protection scope of the present application.

[0043] Figure 1 Flowchart of the AI learning-based advertisement placement strategy optimization method in some embodiments of the present application;

[0044] Figure 2 Flowchart of step S100 of the AI learning-based advertisement placement strategy optimization method in some embodiments of the present application;

[0045] Figure 3 Flowchart of step S300 of the AI learning-based advertisement placement strategy optimization method in some embodiments of the present application;

[0046] Figure 4 Flowchart of step S400 of the AI learning-based advertisement placement strategy optimization method in some embodiments of the present application;

[0047] Figure 5 Hardware structure schematic diagram of the advertisement placement strategy optimization system in some embodiments of the present application.

[0048] The implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0049] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments only represent some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort belong to the protection scope of the present application.

[0050] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present application are only used to explain the relative positional relationship, movement condition, etc. between components in a certain posture (as shown in the drawings), and if the certain posture changes, the directional indications also change accordingly.

[0051] In addition, the description related to "first", "second" and the like in the present application is only for the purpose of description, and cannot be understood as indicating or implying the relative importance of the indicated technical features or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first", "second" can explicitly or implicitly include at least one of the features. In addition, "and / or" throughout the text includes three schemes, for example, A and / or B includes A technical scheme, B technical scheme, and A and B simultaneously meet the technical scheme; in addition, the technical schemes of various embodiments can be combined with each other, but it must be based on the realization of ordinary skilled in the art, when the combination of technical schemes appears contradictory or cannot be realized, it should be considered that the combination of such technical schemes does not exist, nor in the protection scope required by the present application.

[0052] In the era of digital marketing, the precision and efficiency of advertising placement directly affect the operational efficiency of commercial entities. Traditional mall advertising placement strategies usually arrange content based on fixed time periods, preset rules or manual experience, which is difficult to dynamically adapt to environmental changes and real-time fluctuations in user behavior. For example, seasonal changes, weather changes or unexpected events may cause significant changes in consumer shopping preferences, while static advertising strategies cannot capture such environmental relevance in a timely manner, resulting in resource waste and low conversion rates. At the same time, the rapid development of online consumer behavior makes user shopping decisions exhibit "online and offline integration" characteristics, but the offline advertising system lacks the ability to integrate real-time online shopping trends, resulting in insufficient matching of advertising content and user potential needs.

[0053] In the management of shopping malls, environmental conditions have a non-negligible impact on product sales. Different environmental conditions (such as indoor, outdoor environments or changes in corresponding environments) directly or indirectly affect consumer behavior and demand, and thus affect product sales performance. In order to achieve precise optimization of advertising placement and improve the efficiency of mall operations and sales performance, the present application takes the environmental state parameters collected by the environmental sensor in real time as the starting point to provide data support for the subsequent development of advertising strategies.

[0054] As shown in Figures 1-4 The following describes an advertising placement strategy optimization system executing the AI learning-based advertising placement strategy optimization method, it should be noted that although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that here. Please refer to the accompanying drawings Figure 1 The method includes the following steps S100-S600:

[0055] Step S100, acquiring environmental state parameters collected by an environmental sensor in real time, the environmental state parameters including at least one of indoor environment parameters and outdoor environment parameters;

[0056] It can be understood that the environmental state of the mall indoor or outdoor is different, or in other words, the environmental state of the mall indoor or outdoor changes, which has different effects on product sales in the mall. For example, in the hot weather with high outdoor temperature, if the mall can focus on the advertising of products related to "dealing with the heat", such as cool drinks, sunscreens, portable small fans, etc., it will be more in line with the actual needs of consumers at the moment, thereby effectively attracting the attention of customers, stimulating the desire to buy, and thus significantly improving the overall operating efficiency and sales performance of the mall.

[0057] Based on this, the embodiment of the application first acquires the environmental state parameters collected by the environmental sensor in real time, so as to optimize and adjust the advertising according to the environmental state parameters.

[0058] The environmental sensor refers to a sensor for monitoring the environmental state, which can be a people flow detection sensor for monitoring the indoor people flow density, a temperature sensor for detecting the indoor or outdoor environmental temperature, or a rainfall detection sensor for detecting the outdoor rainfall.

[0059] The people flow detection sensor is mainly used for monitoring the indoor people flow density, for example, the people flow density can be detected by image analysis. By real-time sensing of the personnel flow in different areas of the mall, the degree of bustle and consumer gathering in each area can be accurately reflected.

[0060] In an embodiment, the environmental sensor includes an indoor environmental sensor and an outdoor environmental sensor, and step S100: acquiring environmental state parameters collected by the environmental sensor in real time, includes:

[0061] S110, acquiring indoor environmental state parameters collected by the indoor environmental sensor in real time, the indoor environmental state parameters at least including indoor people flow density;

[0062] S120, acquiring outdoor environmental state parameters collected by the outdoor environmental sensor in real time, the outdoor environmental state parameters at least including one of environmental temperature and environmental rainfall.

[0063] Specifically, the indoor environmental sensor (such as the people flow detection sensor) can be distributed in various key areas of the mall, such as the entrance, the elevator port, the main passageway, the periphery of popular stores on each floor, etc. These sensors continuously monitor the indoor environment, and the collected people flow density data is transmitted to the system processor in real time, thereby realizing the detection of the people flow density of the mall.

[0064] The outdoor environment sensor (such as a temperature sensor and a rainfall detection sensor) is installed at a suitable position outside the shopping mall and can accurately perceive the change of the outdoor environment. The temperature sensor monitors the outdoor temperature in real time and transmits the data to the system processor; the rainfall detection sensor is automatically started when it rains, records the rainfall information, and timely uploads the related data.

[0065] The embodiment of the present application sets up a people flow detection sensor indoors to detect the people flow density in the shopping mall, and sets up a temperature sensor or a rainfall sensor outdoors to detect the temperature or rainfall data outside the shopping mall, thereby providing solid and reliable data support and decision basis for subsequent development of more targeted and more consumer demand-oriented advertising delivery strategies.

[0066] For example, the area where a certain shopping mall is located has much rainstorm in summer. When the rainfall detection sensor detects heavy rain, the shopping mall can quickly display the preferential activities of various rain gear through the billboards at the entrance of the shopping mall; at the same time, the rain special entertainment package is promoted in the cinema, game hall and other areas in the shopping mall, so as to guide the consumers to enjoy the fun of shopping and entertainment while avoiding the rain.

[0067] Step S200, obtaining hot online shopping data matched with the environment state parameter according to the environment state parameter;

[0068] After obtaining the environment state parameters inside and outside the shopping mall, in order to further improve the overall operating efficiency and sales performance of the shopping mall, the embodiment of the present application further obtains hot online shopping data matched with the environment state parameters based on the environment state parameters. By combining the environment state with the online consumption trend, the consumer demand is more accurately understood, thereby providing strong support for subsequent advertising delivery strategies.

[0069] In an embodiment, obtaining hot online shopping data matched with the environment state parameter according to the environment state parameter comprises:

[0070] When the indoor people flow density is greater than or equal to the people flow density threshold value and the environment temperature is greater than or equal to the temperature threshold value, a reinforcement learning search algorithm is used to obtain hot-selling commodity data of online shopping in high-temperature weather;

[0071] Or, when the indoor people flow density is greater than or equal to the people flow density threshold value and the environment rainfall is greater than or equal to the rainfall threshold value, a reinforcement learning search algorithm is used to obtain hot-selling commodity data of online shopping in rainy weather.

[0072] In an actual operation scenario, when the indoor people flow density reaches or exceeds the preset people flow density threshold and the outdoor environment temperature reaches or exceeds the preset temperature threshold, it indicates that the shopping mall has a large number of people flow, and the outdoor environment temperature is high. In this case, the shopping demand and consumption preference of consumers are likely to change significantly due to high temperature. To accurately capture this change, the embodiments of the present application introduce a reinforcement learning search algorithm, deeply mine hot-selling commodity data in high-temperature weather from massive data of various online shopping platforms. Taking a hot summer day with continuous high temperature as an example, high temperature will make consumers pay more attention to cooling products, and cool drinks, sunscreens, and light and portable small fans will become their hot choices on the shopping list. Through the reinforcement learning search algorithm, these hot-selling commodities can be efficiently and accurately locked, providing a strong basis for subsequent marketing decisions.

[0073] In another actual operation scenario, when the indoor people flow density reaches or exceeds the preset threshold and the outdoor environment rainfall reaches or exceeds the preset rainfall threshold, it means that there are many consumers in the shopping mall, and it is raining outside. Rainfall will not only affect the travel plans of consumers, but also change their shopping demand. At this time, the online shopping platform is also deeply scanned and searched by means of the reinforcement learning search algorithm to obtain hot-selling commodity data in rainy weather. For example, in rainy days, umbrellas, raincoats, waterproof shoe covers and other rain gear become essential items, and the demand of consumers for them will increase significantly; at the same time, due to the inconvenience of travel, people tend to engage in indoor entertainment activities, so game consoles, books and other indoor entertainment equipment may also become hot-selling commodities.

[0074] In other embodiments, if the indoor people flow density does not exceed the preset people flow density threshold, it indicates that there are few people in the shopping mall, and the optimization of the advertising placement strategy can not be performed, that is, the initial advertising placement strategy can be maintained for advertising placement.

[0075] The reinforcement learning search algorithm in the embodiments of the present application can efficiently screen out hot-selling commodities with high matching degree with environmental state parameters from massive online shopping data by continuously learning and optimizing the search strategy. The algorithm can dynamically adjust the search weight in combination with historical sales data, user behavior data and real-time environmental parameters, and can ensure that the obtained data has both timeliness and high relevance.

[0076] It should be noted that the hot online shopping data can include multiple commodity names matched with the environmental state parameters and data such as the popularity of the corresponding commodities.

[0077] For example, the following is a table of hot-selling commodity data of online shopping in high-temperature weather obtained by the reinforcement learning search algorithm.

[0078]

[0079] Step S300, inputting the environment state parameter and the hot online shopping data into a pre-trained advertisement matching model to output a matching degree score of each candidate advertisement;

[0080] After obtaining the hot online shopping data matched with the environment state parameter, the embodiment of the application inputs the environment state parameter and the hot online shopping data into a pre-trained advertisement matching model to output a matching degree score of each candidate advertisement. In this way, by fusing the environment state parameter and the hot online shopping data, a comprehensive matching degree score of each candidate advertisement is calculated, which provides a basis for subsequent advertisement delivery optimization.

[0081] In an embodiment, step S300: inputting the environment state parameter and the hot online shopping data into a pre-trained advertisement matching model to output a matching degree score of each candidate advertisement, comprises:

[0082] S310, inputting the environment state parameter into a pre-trained advertisement matching model environment subnetwork to obtain an environment attribute score of each candidate advertisement;

[0083] S320, inputting the hot online shopping data into a pre-trained advertisement matching model hot spot subnetwork to obtain a hot spot attribute score of each candidate advertisement;

[0084] S330, performing weighted summation on the environment attribute score and the hot spot attribute score to obtain a matching degree score of each candidate advertisement.

[0085] Specifically, the embodiment of the application first inputs the environment state parameter into a pre-trained advertisement matching model environment subnetwork to obtain an environment attribute score of each candidate advertisement. Then, the hot online shopping data is input into a pre-trained advertisement matching model hot spot subnetwork to obtain a hot spot attribute score of each candidate advertisement. Finally, the environment attribute score and the hot spot attribute score are weighted and summed to obtain a matching degree score of each candidate advertisement.

[0086] Further, if a mall has 10 advertisements to be placed, which are respectively a restaurant A (providing cold drinks), a restaurant B (providing tea drinks), a restaurant C (providing normal temperature water), a daily necessities store A (selling sun-protecting clothes, etc.), a daily necessities store B (selling small fans, umbrellas, etc.), a clothing store A, a clothing store B, a skin care product store A (specializing in selling sunscreens), a skin care product store B (selling a small amount of sunscreen and other skin care products), and a beverage store A. If the environmental state parameter detected by the current environment sensor represents that the outside environment is a high-temperature weather, and the hot online shopping data matched represent that the current hot online shopping products are ranked according to the degree of hot sales as portable fans, sunscreens, and sun-protecting clothes. The environmental state parameter is input into the pre-trained advertisement matching model environment subnetwork to obtain the environmental attribute score of each candidate advertisement, which is the environmental attribute score of each candidate advertisement obtained by model matching according to the environmental state parameter. The higher the matching degree of the environmental state parameter and the candidate advertisement, the higher the score value. For example, when the environmental temperature is high, the advertisements of the stores related to summer cooling (such as stores selling small fans, sunscreens, sun umbrellas, and sun-protecting clothes) have a higher matching degree, while the advertisements of the stores unrelated to high temperature (such as normal temperature drinks and ordinary clothes) have a lower matching degree.

[0087] For example, the scores of the above-mentioned 10 candidate advertisements obtained by model matching according to the environmental state parameter, i.e., the environmental attribute scores, are as follows: the daily necessities store A gets 100 points, the daily necessities store B gets 100 points, the skin care product store A gets 100 points, the beverage store A gets 100 points, the skin care product store B gets 80 points, the restaurant A gets 60 points, the restaurant C gets 40 points, the restaurant B gets 20 points, the clothing store A gets 0 points, and the clothing store B gets 0 points.

[0088] For example, the scores of the above-mentioned 10 candidate advertisements obtained by model matching according to the environmental state parameter, i.e., the environmental attribute scores, are as follows: the daily necessities store A gets 100 points, the daily necessities store B gets 100 points, the skin care product store A gets 100 points, the beverage store A gets 100 points, the skin care product store B gets 80 points, the restaurant A gets 60 points, the restaurant C gets 40 points, the restaurant B gets 20 points, the clothing store A gets 0 points, and the clothing store B gets 0 points.

[0089] After obtaining the environmental attribute scores and the hot attribute scores of each candidate advertisement, the matching degree scores of each candidate advertisement can be obtained by weighted summation according to the preset weights (such as the weights being 0.5).

[0090] According to the environmental state and the hot online shopping data matched with the environmental state, the embodiments of the present application comprehensively score each candidate advertisement to obtain the matching degree scores of each candidate advertisement, thereby providing accurate data support for subsequent optimization of advertisement placement strategies.

[0091] In an embodiment, before the weighting sum of the environment attribute score and the hotspot attribute score is performed to obtain the matching degree score of each candidate advertisement, further comprising: obtaining a change rate of the environment state parameter and a trend acceleration of the hotspot e-commerce data; inputting the change rate of the environment state parameter and the trend acceleration of the hotspot e-commerce data into a dynamic selection network to generate a dynamic weight coefficient, wherein the dynamic selection network satisfies the following expression:

[0092] ;

[0093] ;

[0094] ;

[0095] In the formula, is a dynamic weight of the environment attribute score, is a dynamic weight of the hotspot attribute score; is a change rate of the environment state parameter, is a reference change rate of the environment state parameter, is a reference weight of the environment attribute score; is a trend acceleration of the hotspot e-commerce data, is a reference trend acceleration of the hotspot e-commerce data, is a reference weight of the hotspot attribute score; is a change rate warning value of the environment state parameter, is a trend acceleration warning value of the hotspot e-commerce data.

[0096] Specifically, when the environment state parameter change rate (Φ) significantly increases (exceeds Φ1) and the hotspot trend acceleration H is less than H1, such as when the environment temperature change rate significantly increases but the hotspot trend is not obvious, the weight value α corresponding to the environment attribute score increases, that is, the influence of the environment attribute score is improved; at this time, the dynamic weight of the environment attribute score is calculated according to the reference weight of the environment attribute score. When the hotspot trend acceleration (H) suddenly increases (exceeds H1) and the environment state parameter change rate Φ is less than Φ1, such as when the sales growth rate of the e-commerce commodity matching the environment significantly increases, the weight value β corresponding to the hotspot attribute score increases, that is, the influence of the hotspot attribute score is improved, at this time, the dynamic weight of the hotspot attribute score is calculated according to the reference weight of the hotspot attribute score.

[0097] In this way, the advertisement placement strategy optimization system of the present application can quickly adapt to the sudden changes of the external environment and the e-commerce platform (such as sudden weather changes, holiday promotions, etc.), avoiding the scoring rigidity caused by fixed weights. Moreover, the real-time data driven decision of the embodiment of the present application ensures that the advertisement matching strategy is always dynamically synchronized with the current environment and the e-commerce platform, thereby further improving the accuracy of the advertisement placement strategy optimization.

[0098] Step S400, prioritize each candidate advertisement according to the matching score, generate a first advertisement delivery strategy to preliminarily optimize the initial advertisement delivery strategy;

[0099] After obtaining the matching score of each candidate advertisement with the current environment state and the corresponding hot online shopping data, the preliminary optimization of the advertisement delivery strategy can be performed according to the matching score, such as adjusting the delivery order and delivery time length of the advertisement, etc.

[0100] In an embodiment, step S400: prioritizing each candidate advertisement according to the matching score, generating a first advertisement delivery strategy to preliminarily optimize the initial advertisement delivery strategy, comprises:

[0101] S410, prioritizing each candidate advertisement according to the high-low order of the matching score to obtain an advertisement delivery playlist;

[0102] S420, determining the display time length of each candidate advertisement in the advertisement delivery playlist according to the matching score of each candidate advertisement.

[0103] Specifically, the candidate advertisements can be prioritized according to the high-low order of the matching score to obtain an advertisement delivery playlist, the candidate advertisements with high matching scores are arranged in the front to be displayed preferentially, and the candidate advertisements with low matching scores are arranged in the back. Then, the display time length of each candidate advertisement in the advertisement delivery playlist is determined according to the matching score of each candidate advertisement, the higher the matching score of a candidate advertisement, the higher the matching degree with the environment state, and the longer the display time length. In this way, by adjusting the candidate advertisements with high matching degrees with the environment state to be displayed preferentially and increasing the display time length, the exposure time length of such advertisements is increased, thereby improving the sales conversion of the corresponding goods.

[0104] When determining the display time length of each candidate advertisement in the advertisement delivery playlist according to the matching score of each candidate advertisement, the base display time length of each candidate advertisement can be first determined according to the total advertisement time length constraint and the number of candidate advertisements; then the base display time length is dynamically adjusted according to the matching score of each candidate advertisement to obtain the actual display time length of each candidate advertisement.

[0105] Specifically, in an advertisement total time length constraint period, a base display time length is averagely allocated to the candidate advertisements that need to be displayed. Then, the base display time length is dynamically adjusted according to the matching score of each candidate advertisement to obtain the actual display time length of each candidate advertisement, wherein the display time length can be re-allocated according to the proportion between the matching scores of the candidate advertisements.

[0106] For example, under the time length constraint condition that 10 candidate advertisements are cycled to be displayed for 10 minutes, the benchmark display time length is 1 minute for each candidate advertisement. According to the ratio between the matching degree scores of the candidate advertisements, the embodiment of the application reallocates the display time length, and the specific process is not described herein.

[0107] Thus, the display order of the adjusted candidate advertisements and the corresponding actual display time length are the first advertisement delivery strategy. Therefore, the subsequent playing content of the advertisement delivery screen can be controlled according to the first advertisement delivery strategy, that is, the advertisement delivery on the advertisement delivery screen is performed according to the first advertisement delivery strategy.

[0108] Step S500, controlling the playing content of the advertisement delivery screen according to the first advertisement delivery strategy, and collecting user gaze time length data in real time during the playing process;

[0109] After the advertisement delivery on the advertisement delivery screen according to the first advertisement delivery strategy, the user gaze time length data of each candidate advertisement is collected in real time. The gaze time length data can be the cumulative sum of the gaze time length of all users in one display period, can be the average value of the gaze time length of all users in one display period, or can be the average value of the gaze time length of all users in continuous multiple display periods.

[0110] Step S600, re-optimizing the first advertisement delivery strategy according to the user gaze time length data to obtain a second advertisement delivery strategy.

[0111] Specifically, the delivery efficiency value of each candidate advertisement can be first determined according to the user gaze time length data and the actual display time length of each candidate advertisement, wherein the delivery efficiency value is the ratio of the average value of the user gaze time length to the actual display time length. Then, the first advertisement delivery strategy is re-optimized according to the delivery efficiency value of each candidate advertisement to obtain a second advertisement delivery strategy. For example, when the delivery efficiency value of the candidate advertisement exceeds 1.2 times of the average value of the same type of advertisement in continuous N monitoring periods, a gradual display time length gain strategy is triggered; when the delivery efficiency value of the candidate advertisement is lower than 0.8 times of the average value of the same type of advertisement in continuous M monitoring periods, a display time length hierarchical attenuation strategy is triggered, wherein 5 > N > M > 3.

[0112] For example, if candidate ad A has an average user attention duration of 1 minute and an actual display time of 3 minutes within a campaign period, its campaign efficiency value is 0.33. The calculation of the campaign efficiency values ​​for other candidate ads is similar and will not be repeated here. If candidate ad A's campaign efficiency value exceeds 1.2 times the average of similar ads within 4 consecutive campaign periods, it indicates that the ad's return on investment is relatively high, triggering a progressive display time gain strategy, such as increasing the display time by 10% and 20% in subsequent campaign periods. If candidate ad A's campaign efficiency value is less than 0.8 times the average of similar ads within 3 consecutive campaign periods, it indicates that the ad's return on investment is relatively low, triggering a tiered display time decay strategy, such as decreasing the display time by 10% and 20% in subsequent campaign periods. In this way, ads that users are more interested in receive more display opportunities, while ads that users are less interested in receive fewer display opportunities, thereby achieving secondary optimization of the ad delivery strategy.

[0113] Based on this, the embodiments of this application perform an initial optimization of the initial advertising delivery strategy according to environmental state parameters and matching trending online shopping data to obtain a first advertising delivery strategy. Then, a second optimization is performed based on the delivery efficiency values ​​of each candidate advertisement to obtain a second advertising delivery strategy. This significantly improves advertising effectiveness and user attention, achieving intelligent and precise optimization of the advertising delivery strategy.

[0114] like Figure 5 As shown, Figure 5 The diagram below shows the hardware structure of an advertising strategy optimization system in some embodiments of this application. The advertising strategy optimization system provided in this application also includes a memory 1000 and a processor 2000. The memory 1000 is used to store computer-readable instructions, and the processor 2000 is used to call the computer-readable instructions to execute the AI ​​learning-based advertising strategy optimization method as described above.

[0115] The processor 2000 is configured to provide computing and control capabilities to control the advertisement delivery strategy optimization system to perform corresponding tasks, for example, to control the advertisement delivery strategy optimization system to perform the AI learning-based advertisement delivery strategy optimization method in any of the above method embodiments. The method includes: acquiring environment state parameters of a target area collected by an environment sensor in real time, the environment state parameters including at least one of indoor environment parameters and outdoor environment parameters; acquiring hot online shopping data matched with the environment state parameters according to the environment state parameters; inputting the environment state parameters and the hot online shopping data into a pre-trained advertisement matching model to output a matching score of each candidate advertisement; prioritizing each candidate advertisement according to the matching score to generate a first advertisement delivery strategy to preliminarily optimize an initial advertisement delivery strategy; controlling the playing content of an advertisement delivery screen according to the first advertisement delivery strategy, and collecting user gaze duration data in real time during the playing process; and re-optimizing the first advertisement delivery strategy according to the user gaze duration data to obtain a second advertisement delivery strategy.

[0116] The processor 2000 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a hardware chip, or any combination thereof; and can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The above-mentioned PLD can be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.

[0117] The memory 1000, as a non-transitory computer readable storage medium, can be used to store non-transitory software programs, non-transitory computer executable programs and modules, such as program instructions / modules corresponding to the AI learning-based advertisement delivery strategy optimization method in the embodiments of the present application. The processor 2000 can implement the AI learning-based advertisement delivery strategy optimization method in any of the above method embodiments by running the non-transitory software programs, instructions and modules stored in the memory 1000.

[0118] In particular, the memory 1000 can include volatile memory (VM), such as random access memory (RAM); the memory 1000 can also include non-volatile memory (NVM), such as read-only memory (ROM), flash memory, a hard disk drive (HDD), or a solid-state drive (SSD), or other non-transitory solid-state storage device; the memory 1000 can also include a combination of the above-mentioned types of memory.

[0119] To sum up, the advertisement delivery strategy optimization system of the present application adopts the technical solutions of any one of the above-mentioned advertisement delivery strategy optimization methods based on AI learning, and therefore has at least the beneficial effects brought by the technical solutions of the above-mentioned embodiments, which will not be repeated here.

[0120] The present application also provides a computer readable storage medium, such as a memory including program code, which can be executed by a processor to complete the above-mentioned advertisement delivery strategy optimization method based on AI learning. For example, the computer readable storage medium can be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CDROM), a magnetic tape, a floppy disk, and an optical data storage device, etc.

[0121] The present application also provides a computer program product, which includes one or more program codes stored in a computer readable storage medium. The processor of the early warning system reads the program code from the computer readable storage medium, and the processor executes the program code to complete the steps of the above-mentioned advertisement delivery strategy optimization method based on AI learning.

[0122] Those of ordinary skill in the art can understand that all or part of the steps of the above-mentioned embodiments can be completed by hardware, or by program code related hardware, which can be stored in a computer readable storage medium, such as a read-only memory, a magnetic disk or an optical disk.

[0123] It should be noted that the apparatus embodiments described above are merely illustrative, and the units described as separate units can or can not be physically separate, and the units displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0124] Through the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be implemented by means of software plus a general hardware platform, and of course can also be implemented by hardware. Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware, and the program can be stored in a computer readable storage medium. When the program is executed, it can include the processes of the above-mentioned embodiment methods. The storage medium can be a magnetic disc, an optical disc, a read-only memory (ROM) or a random access memory (RAM), etc.

[0125] The above is only the preferred embodiment of the present application, and does not limit the patent scope of the present application, and any equivalent structural transformation made under the inventive concept of the present application, or direct / indirect application in other related technical fields is included in the patent protection scope of the present application.

Claims

1. An AI learning-based advertisement delivery strategy optimization method applied to an advertisement delivery strategy optimization system, the advertisement delivery strategy optimization system comprising an environment sensor and an advertisement delivery screen, characterized in that, The method comprises: acquiring environment state parameters collected by an environment sensor in real time, the environment state parameters comprising at least one of indoor environment parameters and outdoor environment parameters; acquiring hot online shopping data matched with the environment state parameters according to the environment state parameters; inputting the environment state parameters and the hot online shopping data into a pre-trained advertisement matching model to output matching degree scores of each candidate advertisement; performing priority ranking on each candidate advertisement according to the matching degree scores to generate a first advertisement delivery strategy to preliminarily optimize an initial advertisement delivery strategy; controlling the playing content of an advertisement delivery screen according to the first advertisement delivery strategy, and collecting user gaze duration data in real time during the playing process; optimizing the first advertisement delivery strategy again according to the user gaze duration data to obtain a second advertisement delivery strategy; inputting the environment state parameters and the hot online shopping data into a pre-trained advertisement matching model to output matching degree scores of each candidate advertisement, comprising: inputting the environment state parameters into a pre-trained advertisement matching model environment subnetwork to obtain environment attribute scores of each candidate advertisement; inputting the hot online shopping data into a pre-trained advertisement matching model hot subnetwork to obtain hot attribute scores of each candidate advertisement; performing weighted summation on the environment attribute scores and the hot attribute scores to obtain the matching degree scores of each candidate advertisement; before performing the weighted summation on the environment attribute scores and the hot attribute scores to obtain the matching degree scores of each candidate advertisement, further comprising: acquiring a change rate of the environment state parameters and a trend acceleration of the hot online shopping data; inputting the change rate of the environment state parameters and the trend acceleration of the hot online shopping data into a dynamic selection network to generate a dynamic weight coefficient, wherein the dynamic selection network satisfies the following expression: ； ； ）； wherein, a dynamic weight for the environmental attribute score, a dynamic weight for the hotspot attribute score; a rate of change of the environmental state parameter, a baseline rate of change of the environmental state parameter, a baseline weight for the environmental attribute score; a trend acceleration of the hotspot e-commerce data, a baseline trend acceleration of the hotspot e-commerce data, a baseline weight for the hotspot attribute score; a rate of change of the environmental state parameter alert value, a trend acceleration of the hotspot e-commerce data alert value. 2.The AI learning-based advertisement distribution strategy optimization method of claim 1, wherein, The environment sensor comprises an indoor environment sensor and an outdoor environment sensor, and acquiring the environment state parameters collected by the environment sensor in real time comprises: acquiring indoor environment state parameters collected by the indoor environment sensor in real time, the indoor environment state parameters comprising at least an indoor crowd density; acquiring outdoor environment state parameters collected by the outdoor environment sensor in real time, the outdoor environment state parameters comprising at least one of an environment temperature and an environment rainfall. 3.The AI learning-based advertisement distribution strategy optimization method of claim 2, wherein, Acquiring hot online shopping data matched with the environment state parameters according to the environment state parameters comprises: when the indoor crowd density is greater than or equal to a crowd density threshold value and the environment temperature is greater than or equal to a temperature threshold value, using a reinforcement learning search algorithm to acquire hot-selling commodity data of online shopping in high-temperature weather; or, when the indoor crowd density is greater than or equal to the crowd density threshold value and the environment rainfall is greater than or equal to a rainfall threshold value, using the reinforcement learning search algorithm to acquire hot-selling commodity data of online shopping in rainy weather. 4.The AI learning-based advertisement distribution strategy optimization method of claim 1, wherein, Performing priority ranking on each candidate advertisement according to the matching degree scores to generate a first advertisement delivery strategy to preliminarily optimize an initial advertisement delivery strategy comprises: performing priority ranking on each candidate advertisement according to the high-low order of the matching degree scores to obtain an advertisement delivery playlist; The display time length of each candidate advertisement in the advertisement delivery playlist is determined according to the matching degree score of each candidate advertisement. 5.The AI learning-based advertisement delivery strategy optimization method of claim 4, wherein, The display time length of each candidate advertisement in the advertisement delivery playlist is determined according to the matching degree score of each candidate advertisement, comprising: The reference display time length of each candidate advertisement is determined according to the total time length constraint and the number of candidate advertisements; The actual display time length of each candidate advertisement is dynamically adjusted according to the matching degree score of each candidate advertisement. 6.The AI learning-based advertisement delivery strategy optimization method of claim 5, wherein, The first advertisement delivery strategy is re-optimized according to the user gaze time length data to obtain a second advertisement delivery strategy, comprising: The delivery efficiency value of each candidate advertisement is determined according to the user gaze time length data and the actual display time length of each candidate advertisement, wherein the delivery efficiency value is the ratio of the user gaze time length average value to the actual display time length; The first advertisement delivery strategy is re-optimized according to the delivery efficiency value of each candidate advertisement to obtain a second advertisement delivery strategy. 7.The AI learning-based advertisement distribution strategy optimization method of claim 6, wherein, The first advertisement delivery strategy is re-optimized according to the delivery efficiency value of each candidate advertisement to obtain a second advertisement delivery strategy, comprising: When the delivery efficiency value of the candidate advertisement exceeds 1.2 times of the average value of the same type of advertisement in the continuous N monitoring periods, the gradual display time length gain strategy is triggered; When the delivery efficiency value of the candidate advertisement is lower than 0.8 times of the average value of the same type of advertisement in the continuous M monitoring periods, the display time length grading attenuation strategy is triggered, wherein 5>N>M>3. 8.The AI learning-based advertisement distribution strategy optimization method of claim 1, wherein, The hot spot online shopping data includes a plurality of commodity names matched with the environmental state parameters and the hot degree of the corresponding commodities.

9. An advertisement placement strategy optimization system, comprising: Comprising: A memory and a processor, the memory is used to store program code; The processor is used to call the program code to execute the method of any one of claims 1 to 8.

Images