A Smart Advertising Delivery System and Method Based on User Behavior Analysis

Abstract

This invention relates to the field of artificial intelligence technology, and in particular to an intelligent advertising delivery system and method based on user behavior analysis. The system includes a data acquisition module, a behavior analysis module, an advertising delivery module, and a monitoring and feedback module. The data acquisition module is used to collect target advertising data. The behavior analysis module is used to construct user behavior analysis models and cold start models. The advertising delivery module is used to generate and optimize advertising delivery strategies. The monitoring and feedback module is used to optimize strategies and models based on advertising return on investment. This system improves the accuracy, efficiency, and profitability of advertising delivery through multi-module collaboration, adapts to cold start scenarios, and enhances market adaptability.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an intelligent advertising delivery system and method based on user behavior analysis. Background Technology

[0002] In the field of advertising, the cold start problem has always been a major challenge for advertising professionals. For new users, the platform has no records of their past browsing, purchasing, and other behaviors, rendering traditional advertising strategies based on user behavior analysis ineffective. In new advertising scenarios, because there is no historical data on the ad, it is impossible to know its appeal and conversion rate among different user groups, resulting in a waste of advertising resources. New platforms also face the cold start dilemma. New advertising platforms may lack a mature user behavior data accumulation and analysis system, and cannot effectively utilize user behavior information to optimize advertising. The cold start problem in advertising severely restricts the effectiveness and efficiency of advertising, and an effective solution is urgently needed.

[0003] Chinese Patent Publication No. CN113465672A discloses an intelligent advertising delivery system, including an advertising space detection device. However, this solution still lacks a user behavior data accumulation and analysis system, resulting in insufficient accuracy and low efficiency in advertising delivery. Summary of the Invention

[0004] To address this issue, the present invention provides an intelligent advertising delivery system and method based on user behavior analysis, which overcomes the problems of insufficient accuracy and low efficiency in advertising delivery caused by the lack of user behavior data accumulation and analysis system in the prior art.

[0005] To achieve the above objectives, in one aspect, the present invention provides an intelligent advertising delivery system based on user behavior analysis, comprising:

[0006] The data acquisition module is used to collect target advertising data;

[0007] The behavior analysis module is used to obtain the target comprehensive feature vector based on the target advertising data, to build the user behavior analysis model using the user behavior analysis model building method, to build the cold start model using the cold start model building method, and to optimize the user behavior analysis model based on the cold start model.

[0008] The advertising delivery module is used to generate advertising delivery strategies based on user behavior analysis models through the advertising delivery strategy generation method, and also to adjust the advertising delivery strategies in real time.

[0009] The monitoring and feedback module is used to calculate the return on advertising investment (ROI), obtain the ROI, and optimize the advertising strategy based on the ROI.

[0010] Furthermore, the data acquisition module collects target advertising data, which includes e-commerce data and social media data. Specifically, e-commerce data is collected by embedding code points on e-commerce websites and mobile applications, and social media data is collected through application programming interfaces (APIs). The e-commerce data includes the number of views, clicks, purchases, the price of purchased goods, the purchase time interval, and the user's browsing time on the platform. The social media data includes user comments, user reposts, user likes, user shares, the number of followers in the user's social circle, and historical advertising records on the e-commerce platform.

[0011] Furthermore, the behavior analysis module obtains the target comprehensive feature vector based on the target advertising data, the process of which includes:

[0012] Step J01: Calculate the browsing frequency d1 according to the formula d1=a1 / t, calculate the click frequency d2 according to the formula d2=a2 / t, and calculate the purchase frequency d3 according to the formula d3=a3 / t, where a1 is the number of views, a2 is the number of clicks, a3 is the number of purchases, and t is the preset duration. Use the browsing frequency d1, click frequency d2, and purchase frequency d3 as behavioral frequency features.

[0013] Step J02: Obtain product association features according to product association rules;

[0014] Step J03: Compare the price h of the purchased item with the average price h0 of similar items. Based on the comparison result, determine the price sensitivity of the purchased item h, and output the price sensitivity based on the determination result.

[0015] When h < h0, the price h of the purchased goods is judged to be too high, and the price sensitivity output is low. Low sensitivity is used as the price sensitivity feature.

[0016] When h≥h0, the price h of the purchased goods is determined to be low, and the price sensitivity output is high, and high sensitivity is used as the price sensitivity feature.

[0017] Step J04: Use behavioral frequency features, product association features, and price sensitivity features as e-commerce data features;

[0018] Step J05: Perform sentiment analysis on user comments using natural language processing to obtain sentiment tendency features; perform interest analysis on user reposts, likes, and shares using natural language processing to obtain interest topic features; perform social analysis on user shares and the number of followers in the user's social circle using natural language processing to obtain social influence features; and use sentiment tendency features, interest topic features, and social influence features as social media data features.

[0019] Step J06: Perform data normalization processing on the e-commerce data features and social media data features to obtain the target comprehensive feature vector.

[0020] Furthermore, the behavior analysis module constructs a user behavior analysis model using a user behavior analysis model construction method, which includes:

[0021] Step U01: Divide the historical comprehensive database into a 70% user behavior training set, a 20% user behavior validation set, and a 10% user behavior test set;

[0022] Step U02: Set the number of weak learners in the gradient boosting decision tree model parameters to 100 and the learning rate to 0.1. Train the gradient boosting decision tree model with the user behavior training set input parameters to obtain the trained gradient boosting decision tree model.

[0023] Step U03: Input the user behavior validation set into the trained gradient boosting decision tree model, optimize the parameters of the trained gradient boosting decision tree model to obtain the parameter-optimized gradient boosting decision tree model, and then input the user behavior test set into the parameter-optimized gradient boosting decision tree model for testing, and output the test accuracy R.

[0024] Furthermore, the user behavior analysis model construction method also includes: step U04, comparing the test accuracy R with the preset test accuracy R0, judging the compliance status of the parameter-optimized gradient boosting decision tree model based on the comparison result, and outputting the parameter-optimized gradient boosting decision tree model based on the judgment result.

[0025] Furthermore, the behavior analysis module constructs a cold start model using a cold start model construction method, which includes:

[0026] Step Q01 involves comparing the preset ad placements with the historical ad data search database. Based on the comparison results, the existence of the preset ad placements in the historical ad data search database is determined, and the definition of the preset ad placements is output based on the determination result. Specifically:

[0027] When the preset ad does not exist in the historical ad data search database, the definition of the preset ad will be output as a new ad.

[0028] When the existence of the preset ad in the historical ad data search database is found to be "existing", the definition of the preset ad will be output as the old ad.

[0029] Step Q02 involves comparing the preset advertising e-commerce platform with the historical advertising e-commerce platform database. Based on the comparison results, the existence of the preset advertising e-commerce platform in the historical advertising e-commerce platform database is determined, and the definition of the preset advertising e-commerce platform is output based on the determination result. Specifically:

[0030] When the preset advertising e-commerce platform does not exist in the historical advertising e-commerce platform database, the definition of the preset advertising e-commerce platform will be output as a new platform.

[0031] When the existence of the preset advertising e-commerce platform in the historical advertising e-commerce platform database is found to be true, the definition of the preset advertising e-commerce platform will be output as the old platform.

[0032] Step Q03: Compare the user's browsing time t1 within a preset time period with the preset browsing time t0. Based on the comparison result, determine the attributes of the user's browsing time t1 within the preset time period, and output the user type based on the determination result.

[0033] Step Q04: Divide 70% of the target comprehensive feature vector into a cold start training set and 30% of the target comprehensive feature vector into a cold start validation set.

[0034] Step Q05: Select the prototype network algorithm as the cold start model, initialize the weights and biases of the cold start model, input the cold start training set into the cold start model, and calculate the output of the cold start model.

[0035] Step Q06: Calculate the loss function value based on the output and label of the cold start model, calculate the gradient through the backpropagation algorithm, and update the weights and biases of the cold start model. Repeat the forward propagation, loss function calculation and backpropagation process to obtain the trained cold start model.

[0036] Step Q07: Input the cold start validation set into the trained cold start model for testing, and output the trained cold start model with an accuracy of 90% as the cold start model.

[0037] Furthermore, the behavior analysis module obtains the user's platform browsing time t1 within a preset time t2, calculates the user's like frequency according to the formula Fd = f / t2, where f is the number of likes by the user, calculates the user's comment frequency FP according to the formula FP = f2 / t2, where f2 is the number of comments by the user, and calculates the matching degree P between the user group and the platform according to the formula P = ω1×t1 + ω2×Fd + ω3×FP, where ω1 is the weight parameter of the user's platform browsing time t1 within the preset time t2, ω2 is the weight parameter of the user's like frequency, and ω3 is the weight parameter of the user's comment frequency. The matching degree P between the user group and the platform is compared with the preset matching degree P0, and the degree of matching between the user group and the platform is judged according to the comparison result. The cold start model is then adjusted according to the judgment result.

[0038] Furthermore, the advertising delivery module generates an advertising delivery strategy using an advertising delivery strategy generation method, which includes:

[0039] Step K01: The probability G of a user purchasing a product is obtained through a user behavior analysis model and compared with various preset purchase probabilities, including a first preset purchase probability G1 and a second preset purchase probability G2. Based on the comparison results, the likelihood of the user purchasing the product is judged, and the user's purchase attributes are output based on the judgment results.

[0040] When G≤G1, the advertising delivery module determines that the probability G of a user purchasing a product is low and outputs the user's purchase attribute as a user with low purchase probability.

[0041] When G1 < G ≤ G2, the advertising delivery module determines the probability G of a user purchasing a product as medium probability and outputs the user's purchase attribute as a user with medium purchase probability.

[0042] When G > G2, the advertising delivery module determines that the probability G of a user purchasing a product is high and outputs the user's purchase attribute as a user with a high probability of purchasing.

[0043] Step K02: When the user's purchase attribute output is "low purchase probability user", the advertising strategy output is: N1 times within a preset period, the advertising content is popular products and newly launched products, providing product information and purchase guidance;

[0044] When the user's purchase attribute output is "medium purchase probability user", the advertising strategy output will be: N2 times within a preset period, the advertising content will be the value and practicality of the product, providing product information and purchase guidance;

[0045] When the user's purchase attribute output is "high purchase probability user", the advertising strategy output will be: N3 times within a preset period, and the advertising content will be the advantages and features of the product, providing product information and purchase guidance;

[0046] The advertising delivery module compares the purchase time interval TG with the preset purchase time interval TG0, judges the key attributes of the purchase time interval TG based on the comparison results, and adjusts the advertising delivery strategy based on the judgment results.

[0047] When the advertising delivery module adjusts the advertising delivery strategy, for users with low purchase probability, the adjusted number of deliveries is set to N1`, N1` = N1. After a preset delivery duration, the probability G of the user purchasing the product is updated, and the purchase update coefficient is set to β, 1.1 ≤ β ≤ 1.5. The updated probability of the user purchasing the product is G`, G` = β × G.

[0048] For users with a medium probability of purchasing, the adjusted number of deliveries is set to N2`, where N2` = 1.2 × N2;

[0049] For users with a high probability of purchasing, the adjusted number of deliveries is set to N3`, where N3` = 2.5 × N3;

[0050] The advertising module obtains the product's sales growth rate M1, inventory level M2, and competitor product sales growth rate M3 from market research reports. It calculates the product promotion index C using the formula C = 0.4 × M1 + 0.2 × M2 + 0.4 × M3, compares it with a preset promotion index C0, judges the impact of the comparison on the product promotion index C based on the comparison results, and updates the purchase time interval TG based on the judgment results.

[0051] When C≤C0, the advertising delivery module determines that the influence of the product promotion index C is low and does not update the purchase time interval TG.

[0052] When C > C0, the advertising delivery module determines that the influence of the product promotion index C is high and updates the purchase time interval TG. For users with low purchase probability, N1` = N2` = N3` is set; for users with medium purchase probability, N1` = N2` = N3` is set; for users with high purchase probability, no adjustment is made to the advertising delivery strategy.

[0053] Furthermore, the monitoring and feedback module obtains the product investment cost I0 and the product investment return I based on market research institution reports, and then uses the formula... The Return on Investment (ROI) for advertising is calculated and compared with a preset ROI of 0. Based on the comparison results, the return attribute of the ROI is determined, and the advertising strategy is optimized accordingly.

[0054] When ROI≥ROI0, the monitoring and feedback module determines that the return attribute of the advertising investment return rate ROI is high and does not optimize the advertising placement strategy.

[0055] When ROI < ROI 0, the monitoring and feedback module determines that the return attribute of the advertising investment ROI is low and optimizes the advertising placement strategy accordingly.

[0056] For users with a low probability of purchase, the first optimization coefficient is set to θ1, where θ1 = 1.7 - 3.2e -1.1×[(ROIO-ROI)-2 The optimized number of deliveries is N1``, N1``=θ1×N1`;

[0057] For users with a medium probability of purchasing, the second optimization coefficient is set to θ2, where θ2 = 1.2 - 3.3e. -1.2×[(ROIO-ROI)+1] The optimized number of deliveries is N2, where N2 = θ2 × N2.

[0058] For users with a high probability of purchase, the third optimization coefficient is set to θ3, where θ3 = 1.5 - 3.2e -1.2×[(ROIO-ROI)+2] The optimized number of deliveries is N3``, N3``=θ3×N3`.

[0059] On the other hand, the present invention also provides an intelligent advertising delivery method based on user behavior analysis, comprising:

[0060] Step S01 is used to collect target advertising data;

[0061] Step S02 is used to obtain the target comprehensive feature vector based on the target advertising data, to construct the user behavior analysis model using the user behavior analysis model construction method, to construct the cold start model using the cold start model construction method, and to optimize the user behavior analysis model based on the cold start model.

[0062] Step S03 is used to generate an advertising strategy based on a user behavior analysis model using an advertising strategy generation method, and also to adjust the advertising strategy in real time.

[0063] Step S04 is used to calculate the return on advertising investment (ROI), obtain the ROI, and optimize the advertising strategy based on the ROI.

[0064] Compared with the prior art, the beneficial effect of the present invention is that, through the efficient collaborative operation of multiple modules, a refined management and analysis system covering intelligent advertising placement is constructed, so as to analyze the duration of advertising placement from multiple dimensions and effectively improve the accuracy and efficiency of advertising placement. Attached Figure Description

[0065] Figure 1 This is a schematic diagram of the intelligent advertising delivery system based on user behavior analysis in this embodiment;

[0066] Figure 2 This is a flowchart illustrating the intelligent advertising delivery method based on user behavior analysis in this embodiment. Detailed Implementation

[0067] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0068] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0069] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0070] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0071] Please see Figure 1 As shown, this is a schematic diagram of the intelligent advertising delivery system based on user behavior analysis in this embodiment. The system includes:

[0072] The data acquisition module is used to collect target advertising data;

[0073] The behavior analysis module is used to obtain the target comprehensive feature vector based on the target advertising data, to build the user behavior analysis model using the user behavior analysis model building method, to build the cold start model using the cold start model building method, and to optimize the user behavior analysis model based on the cold start model. The behavior analysis module is connected to the data acquisition module.

[0074] The ad delivery module is used to generate ad delivery strategies based on user behavior analysis models using ad delivery strategy generation methods. It is also used to adjust ad delivery strategies in real time. The ad delivery module is connected to the behavior analysis module.

[0075] The monitoring and feedback module is used to calculate the return on advertising investment (ROI), obtain the ROI, and optimize the advertising strategy based on the ROI. The monitoring and feedback module is connected to the advertising delivery module.

[0076] Specifically, the system is applied to the terminal of an advertising delivery system. Through the collaborative operation of multiple modules, it performs refined management of the entire intelligent advertising delivery process, thereby improving the accuracy, efficiency, and profitability of advertising delivery. Specifically, the system collects e-commerce and social media data through code embedding and application programming interfaces in the data acquisition module, improving the comprehensiveness and real-time nature of data collection. The system constructs user behavior analysis and cold start models through the behavior analysis module, combining target comprehensive feature vector analysis to determine the matching degree between user behavior and the platform, improving the accuracy of user behavior analysis and the adaptability of the cold start model. The system generates advertising delivery strategies through probability classification of user product purchases in the advertising delivery module, and dynamically adjusts the strategies based on purchase time intervals and product promotion indices, improving the targeting and flexibility of advertising delivery. The system calculates the return on investment (ROI) of advertising through the monitoring and feedback module to optimize advertising delivery strategies, improving the economic benefits and model performance of advertising delivery.

[0077] Specifically, the data acquisition module collects target advertising data, which includes e-commerce data and social media data. The e-commerce data includes pageview counts, click counts, purchase counts, purchase prices, purchase time intervals, and user browsing time on the platform. This embodiment does not limit the method of collecting e-commerce data; those skilled in the art can freely configure it according to actual conditions, as long as it meets the requirement of collecting user purchase characteristics on the e-commerce platform. For example, in this embodiment, the data acquisition module collects e-commerce data by embedding code points in e-commerce websites and mobile applications. The social media data includes user comments, user reposts, user likes, user shares, and the number of followers in a user's social circle. This embodiment does not limit the method of collecting social media data; those skilled in the art can freely configure it according to actual conditions, as long as it meets the requirement of collecting user behavioral characteristics on the social media platform. For example, in this embodiment, the data acquisition module collects social media data through an application programming interface (API).

[0078] Specifically, the e-commerce data refers to data generated when users engage in shopping-related activities on e-commerce platforms, reflecting their consumption behavior. The social media data refers to user interaction data on social media platforms, reflecting their social behavior. The e-commerce website refers to a website used for online goods transactions. The mobile application refers to an e-commerce application installed on a mobile device. Code tracking refers to inserting specific statistical code into the code of the e-commerce website and mobile application. This specific statistical code is code used to track user behavior data, such as JavaScript tags. The application programming interface (API) refers to the standardized interface provided by the social media platform, through which social media data can be obtained after authorization. The number of views refers to the total number of times a user visits the e-commerce platform. The number of clicks refers to the total number of times a user clicks on interactive entry points on the e-commerce platform. This embodiment does not limit the interactive entry points, such as product images and purchase buttons. The number of purchases refers to the number of times a user clicks on an interactive entry point on the e-commerce platform. The total number of times a user completes an order and payment on an e-commerce platform, where the purchase time interval refers to the time difference between two consecutive purchases, the purchase price refers to the actual amount paid by the user for a single item, the user browsing time refers to the total duration of the user's visit to the e-commerce platform, the user's comments refer to the text content posted by the user on social media regarding advertisements and products, including the number of comments, the user's reposts refer to the user sharing advertisement content and product links to their social circle, the user's likes refer to the number of clicks by the user indicating approval of the advertisement and product, including the number of likes, the user's sharing refers to the user's act of publicly disseminating products from the e-commerce platform through social media, and the number of followers in the user's social circle refers to the total number of followers the user has on social media. This embodiment does not limit the types of social media mentioned; those skilled in the art can freely set them according to actual circumstances, as long as they meet the user's needs for online social interaction, such as Weibo.

[0079] Specifically, the data acquisition module collects e-commerce and social media data through code embedding and application programming interfaces, thereby improving the comprehensiveness and real-time performance of data collection.

[0080] Specifically, the behavior analysis module obtains the target comprehensive feature vector based on the target advertising data, and the process includes:

[0081] Step J01: Calculate the browsing frequency d1 according to the formula d1=a1 / t, calculate the click frequency d2 according to the formula d2=a2 / t, and calculate the purchase frequency d3 according to the formula d3=a3 / t, where a1 is the number of views, a2 is the number of clicks, a3 is the number of purchases, and t is the preset duration. Use the browsing frequency d1, click frequency d2, and purchase frequency d3 as behavioral frequency features.

[0082] Step J02: Obtain product association features according to product association rules;

[0083] Step J03: Compare the price h of the purchased item with the average price h0 of similar items. Based on the comparison result, determine the price sensitivity of the purchased item h, and output the price sensitivity based on the determination result.

[0084] When h < h0, the price h of the purchased goods is judged to be too high, and the price sensitivity output is low. Low sensitivity is used as the price sensitivity feature.

[0085] When h≥h0, the price h of the purchased goods is determined to be low, and the price sensitivity output is high, and high sensitivity is used as the price sensitivity feature.

[0086] Step J04: Use behavioral frequency features, product association features, and price sensitivity features as e-commerce data features;

[0087] Step J05: Perform sentiment analysis on user comments using natural language processing to obtain sentiment tendency features; perform interest analysis on user reposts, likes, and shares using natural language processing to obtain interest topic features; perform social analysis on user shares and the number of followers in the user's social circle using natural language processing to obtain social influence features; and use sentiment tendency features, interest topic features, and social influence features as social media data features.

[0088] Step J06: Perform data normalization processing on the e-commerce data features and social media data features to obtain the target comprehensive feature vector.

[0089] Specifically, the product association rules refer to a data analysis method that discovers the association characteristics of different products during the sales process by mining sales data and user behavior data. In this embodiment, the product association rules are set to group similar products and customers to discover the association between products within the group. For example, products with similar functions, price ranges, and target customer groups are grouped into one category. The sales data refers to the core data of users completing transactions on the e-commerce platform, such as product category, unit price, and sales volume. The user behavior data refers to the behavioral records generated by users during their interaction with the system and platform, such as browsing page content and browsing time. The product association characteristics refer to the features reflecting the relationship between products obtained based on the product association rule analysis. This embodiment does not specifically limit the form of product association characteristics. Those skilled in the art can freely choose according to actual needs, as long as the need to obtain product association characteristics is met, such as product group association. The average price of similar products refers to the price value obtained by statistically averaging the prices of a group of products with similar attributes, functions, and uses. The price sensitivity refers to the degree to which consumers react to changes in product prices. The natural language processing refers to the technical means of enabling computers to classify, analyze, and extract information from data, and to understand, process, and generate human language. This embodiment does not limit the specific methods of natural language processing. Those skilled in the art can freely choose according to actual needs, as long as the requirements of obtaining sentiment characteristics and interest topic characteristics are met. For example, a keyword extraction method can be set to obtain sentiment characteristics, and a semantic understanding method can be set to obtain interest topic characteristics. The sentiment analysis mentioned refers to an application of natural language processing, which obtains the sentiment tendency of the text by analyzing the vocabulary, sentence structure, etc. expressed in the user's comments. This embodiment does not limit the sentiment tendency of the text, such as positive, negative, and neutral. The interest analysis mentioned refers to the analysis of user forwards and user likes through natural language processing. The analysis of user-shared content identifies topics of interest to users. This embodiment does not limit the topics of interest to users, such as fashion, technology, or food. Social analysis refers to using natural language processing to acquire user influence on social networks. This embodiment does not limit user influence on social networks, such as user shares or the number of followers in a user's social circle. Data normalization refers to converting data to a uniform scale to eliminate differences in numerical magnitude and units of different features, making the data more suitable for data analysis. This embodiment does not limit data normalization to linear normalization.

[0090] Specifically, the behavior analysis module improves the accuracy of ad placement by extracting and normalizing features from e-commerce and social media data.

[0091] Specifically, the behavior analysis module constructs a user behavior analysis model using a user behavior analysis model construction method, which includes:

[0092] Step U01: Divide the historical comprehensive database into a 70% user behavior training set, a 20% user behavior validation set, and a 10% user behavior test set;

[0093] Step U02: Set the number of weak learners in the gradient boosting decision tree model parameters to 100 and the learning rate to 0.1. Train the gradient boosting decision tree model with the user behavior training set input parameters to obtain the trained gradient boosting decision tree model.

[0094] Step U03: Input the user behavior verification set into the trained gradient boosting decision tree model, optimize the parameters of the trained gradient boosting decision tree model to obtain the parameter-optimized gradient boosting decision tree model, and then input the user behavior test set into the parameter-optimized gradient boosting decision tree model for testing, and output the test accuracy R.

[0095] Step U04: Compare the test accuracy R with the preset test accuracy R0. Based on the comparison result, determine the compliance status of the parameter-optimized gradient boosting decision tree model, and output the parameter-optimized gradient boosting decision tree model based on the determination result.

[0096] When R≥R0, the behavior analysis module determines that the parameter-optimized gradient boosting decision tree model meets the standard, and outputs the parameter-optimized gradient boosting decision tree model as the user behavior analysis model.

[0097] When R < R0, the behavior analysis module determines that the gradient boosting decision tree model after parameter optimization is not up to standard, obtains the second e-commerce data features and social media data features, and trains the gradient boosting decision tree model based on the second e-commerce data features and social media data features until the gradient boosting decision tree model after parameter optimization meets the standard, and outputs the gradient boosting decision tree model after parameter optimization as the user behavior analysis model.

[0098] Specifically, the historical comprehensive database includes e-commerce data features and social media data features from historical records as input data for the user behavior analysis model, and the probability of a user purchasing a product as output data for the user behavior analysis model. The user behavior training set refers to the dataset used to train the gradient boosting decision tree model. The user behavior validation set refers to the dataset used to verify the model's performance and evaluate its performance on new data during model training. The user behavior test set refers to the dataset used to test the accuracy of the gradient boosting decision tree model after training and parameter optimization. The number of weak learners refers to the number of gradient boosting decision tree models. The tree model contains the number of simple decision trees. The learning rate refers to the parameter that controls the impact of the number of newly added weak learners on the improvement of the decision tree model during each iteration in the gradient boosting decision tree model training process. The parameter optimization refers to the process of adjusting the model parameters, such as the weights of weak learners and the splitting conditions of decision trees, to minimize the loss function of the model on the validation set and the training set. The loss function is a value used to measure the difference between the model's predicted results and the actual results. The preset test accuracy R0 is a pre-set value used to measure the performance of the gradient boosting decision tree model after parameter optimization. In this embodiment, the test accuracy R0 is set to 90%.

[0099] Specifically, the behavior analysis module adopts a gradient boosting decision tree model, which improves the accuracy and reliability of the user behavior analysis model by scientifically dividing the dataset and optimizing parameters.

[0100] Specifically, the behavior analysis module constructs a cold start model using a cold start model construction method, which includes:

[0101] Step Q01 involves comparing the preset ad placements with the historical ad data search database. Based on the comparison results, the existence of the preset ad placements in the historical ad data search database is determined, and the definition of the preset ad placements is output based on the determination result. Specifically:

[0102] When the preset ad does not exist in the historical ad data search database, the definition of the preset ad will be output as a new ad.

[0103] When the existence of the preset ad in the historical ad data search database is found to be "existing", the definition of the preset ad will be output as the old ad.

[0104] Step Q02 involves comparing the preset advertising e-commerce platform with the historical advertising e-commerce platform database. Based on the comparison results, the existence of the preset advertising e-commerce platform in the historical advertising e-commerce platform database is determined, and the definition of the preset advertising e-commerce platform is output based on the determination result. Specifically:

[0105] When the preset advertising e-commerce platform does not exist in the historical advertising e-commerce platform database, the definition of the preset advertising e-commerce platform will be output as a new platform.

[0106] When the existence of the preset advertising e-commerce platform in the historical advertising e-commerce platform database is found to be true, the definition of the preset advertising e-commerce platform will be output as the old platform.

[0107] Step Q03: Compare the user's platform browsing time t1 within a preset time period with the preset browsing time t0. Based on the comparison result, determine the attributes of the user's platform browsing time t1 within the preset time period, and output the user type based on the determination result, where:

[0108] When t1≥t0, the behavior analysis module determines that the user's browsing time t1 within the preset time is a long time and outputs the user type as old user.

[0109] When t1 < t0, the behavior analysis module determines that the user's browsing time t1 within the preset time is short and outputs the user type as new user.

[0110] Step Q04: Divide 70% of the target comprehensive feature vector into a cold start training set and 30% of the target comprehensive feature vector into a cold start validation set.

[0111] Step Q05: Select the prototype network algorithm as the cold start model, initialize the weights and biases of the cold start model, input the cold start training set into the cold start model, and calculate the output of the cold start model.

[0112] Step Q06: Calculate the loss function value based on the output and label of the cold start model, calculate the gradient through the backpropagation algorithm, and update the weights and biases of the cold start model. Repeat the forward propagation, loss function calculation and backpropagation process to obtain the trained cold start model.

[0113] Step Q07: Input the cold start validation set into the trained cold start model for testing, and output the trained cold start model with an accuracy of 90% as the cold start model.

[0114] Specifically, the preset advertising refers to the advertisement to be placed, and the historical advertising data search library refers to a database storing advertisements that have been placed on the e-commerce platform in the past. The historical advertising data search library stores data in the form of data, such as preset advertising name - historical advertising content. The existence of the preset advertising in the historical advertising data search library means that the preset advertising exists in the historical advertising data search library. This embodiment does not limit the method for determining the existence of the preset advertising in the historical advertising data search library. Those skilled in the art can freely choose according to actual needs, as long as the need to determine the existence of the preset advertising in the historical advertising data search library is met. The system directly searches the historical advertising data database using preset advertising placements to determine if there are any identical records. The preset advertising placement e-commerce platform refers to the e-commerce platform for which the advertising is pre-determined to be placed. The historical advertising e-commerce platform database is a collection of data storing historical advertising e-commerce platform data. This database stores data in data format, such as "Preset Advertising Placement E-commerce Platform Name - Historical Platform Name". The existence of the preset advertising placement e-commerce platform in the historical advertising e-commerce platform database means that the preset advertising placement e-commerce platform exists in the database. This embodiment does not perform a determination of the existence of the preset advertising placement e-commerce platform in the historical advertising e-commerce platform database. The method of judgment is not limited, and those skilled in the art can freely choose according to actual needs. It only needs to meet the requirement of judging the existence of a preset advertising e-commerce platform in the historical advertising e-commerce platform database. For example, it can directly search the historical advertising e-commerce platform database through the preset advertising e-commerce platform to determine whether there are completely identical records. The preset time refers to a pre-set period of time. This embodiment does not limit the preset time; for example, it can be set to 24 hours. The preset browsing time t0 refers to a pre-set time length used to compare with the user's browsing time within the preset time to judge the attributes of the user's browsing time within the preset time. This embodiment does not limit the preset browsing time t0. In this embodiment, t0 is set to 2 hours. The prototype network algorithm refers to achieving fast classification by extracting category prototypes. The category prototype refers to a representative feature vector. In this embodiment, the category prototype is the target comprehensive feature vector. The cold start training set refers to the dataset used to train the cold start model. The cold start validation set refers to the dataset used to validate the performance of the cold start model during training. Initialization refers to the process of setting initial values ​​for the model's weights and biases when building the cold start model. This embodiment does not limit the initialization method. Those skilled in the art can freely choose according to actual needs, as long as the requirement of setting initial values ​​for the model's weights and biases is met, such as random initialization.The loss function value refers to a quantitative indicator that measures the error between the model's predicted result and the true label. In this embodiment, the loss function is set to mean squared error. The label refers to the output result value corresponding to the input value in the cold start model. The accuracy test refers to the process of evaluating the trained cold start model using a cold start validation set to determine the proportion of the cold start model's predicted result that matches the true result.

[0115] Specifically, the behavior analysis module combines the target comprehensive feature vector to analyze the matching degree between user behavior and the platform, thereby improving the accuracy of user behavior analysis and the adaptability of the cold start model.

[0116] Specifically, the behavior analysis module obtains the user's platform browsing time t1 within a preset time t2, calculates the user's like frequency according to the formula Fd = f / t2, where f is the number of likes, calculates the user's comment frequency FP according to the formula FP = f2 / t2, where f2 is the number of comments, and calculates the matching degree P between the user group and the platform according to the formula P = ω1×t1 + ω2×Fd + ω3×FP, where ω1 is the weight parameter of the user's platform browsing time t1 within the preset time t2, ω2 is the weight parameter of the user's like frequency, and ω3 is the weight parameter of the user's comment frequency. The matching degree P between the user group and the platform is compared with a preset matching degree P0. Based on the comparison result, the degree of matching between the user group and the platform is judged, and the cold start model is adjusted according to the judgment result.

[0117] When P≥P0, the behavior analysis module determines that the user group and the platform are highly matched and does not adjust the cold start model.

[0118] When P < P0, the behavior analysis module determines that the matching degree between the user group and the platform is low, and adjusts the cold start model by adding e-commerce data feature vectors and social media data feature vectors of old users whose platform and new user characteristics are similar to those of new users to the target comprehensive feature vector.

[0119] Specifically, the number of likes refers to the total number of times a user performs a like operation on the platform within a preset time t2, where the platform refers to a preset e-commerce platform for advertising. The number of comments refers to the total number of times a user performs a comment operation on the platform within the preset time t2. The weight parameter refers to a preset parameter value used to measure the relative importance of influencing factors in assessing the matching degree between the user group and the platform. The measured influence refers to the user's platform browsing time t1, the frequency of user likes, and the frequency of user comments within the preset time t2. The preset matching degree P0 refers to a preset standard value used to measure the degree of matching between the user group and the platform. This embodiment does not limit the preset matching degree P0; this embodiment sets P0 = 0.7, the new user characteristics refer to the various behaviors, attributes, and interests exhibited by new users on the platform. The old users with similar new user characteristics refer to old users who are similar to new users in terms of behavior, attributes, and interests. The e-commerce data feature vector and social media data feature vector of old users with similar new user characteristics refer to the extracted behavioral data of old users on e-commerce platforms and social media, which are converted into numerical feature vectors to supplement the input information of the cold start model. In this embodiment, by extracting features from the e-commerce data and social media data of old users and normalizing the extracted features, the e-commerce data feature vector and social media data feature vector of old users with similar new user characteristics are obtained.

[0120] Specifically, the behavior analysis module calculates the matching degree P between user groups and the platform through multi-dimensional data, dynamically adjusts the cold start model, and enhances the matching between new users and the platform by leveraging similar old user characteristics, thereby improving the accuracy of the cold start model.

[0121] Specifically, the advertising delivery module generates an advertising delivery strategy using an advertising delivery strategy generation method, which includes:

[0122] Step K01: The probability G of a user purchasing a product is obtained through a user behavior analysis model and compared with various preset purchase probabilities, including a first preset purchase probability G1 and a second preset purchase probability G2. Based on the comparison results, the likelihood of the user purchasing the product is judged, and the user's purchase attributes are output based on the judgment results.

[0123] When G≤G1, the advertising delivery module determines that the probability G of a user purchasing a product is low and outputs the user's purchase attribute as a user with low purchase probability.

[0124] When G1 < G ≤ G2, the advertising delivery module determines the probability G of a user purchasing a product as medium probability and outputs the user's purchase attribute as a user with medium purchase probability.

[0125] When G > G2, the advertising delivery module determines that the probability G of a user purchasing a product is high and outputs the user's purchase attribute as a user with a high probability of purchasing.

[0126] Step K02: When the user's purchase attribute output is "low purchase probability user", the advertising strategy output is: N1 times within a preset period, the advertising content is popular products and newly launched products, providing product information and purchase guidance;

[0127] When the user's purchase attribute output is "medium purchase probability user", the advertising strategy output will be: N2 times within a preset period, the advertising content will be the value and practicality of the product, providing product information and purchase guidance;

[0128] When a user's purchase attribute output is "high purchase probability user", the advertising strategy output will be: to place ads N3 times within a preset period, and the ad content will focus on the advantages and features of the product, providing product information and purchase guidance.

[0129] Specifically, the first preset purchase probability G1 refers to a pre-set lower threshold value used to determine the probability of a user purchasing a product, and the second preset purchase probability G2 refers to a pre-set upper threshold value used to determine the probability of a user purchasing a product. In this embodiment, G1 = 0.3 and G2 = 0.7. The preset period refers to a pre-set time length. In this embodiment, the preset period is not limited; for example, it can be set to one week. The number of ad placements refers to the number of ad placements within the preset period. In this embodiment, the number of placements is not limited to N1; for example, it can be set to N1 = 2 times. The popular products refer to products with high sales and popularity on the platform. In this embodiment, popular products are obtained through online statistics. The online statistics refer to the method of statistically analyzing the click-through rate and purchase rate of products. The newly launched products refer to products that are launched and promoted on the platform within the new product preset period. The new product preset period refers to the length of the most recent period of time from the current moment. In this embodiment, a new product preset period is set. The period is one week. The product information refers to the key features, value, and differentiated advantages of the product conveyed to users through advertising. In this embodiment, information content is matched according to the user's purchase probability. The purchase guidance refers to the operation of prompting users to complete the purchase behavior through copywriting, activities, etc. In this embodiment, the purchase guidance is set as a benefit incentive. In this embodiment, the number of times the advertisement is placed is not limited to N2. For example, if N2 = 4 times, the value of the product refers to the benefits and advantages that the product can bring to the user and the degree to which it meets the user's needs. In this embodiment, the value of the product is obtained through user research. In this embodiment, the method of user research is not limited, such as questionnaire surveys. The practicality refers to the operability and applicability of the product in actual life and work. In this embodiment, practicality content is set through user research, such as long battery life. In this embodiment, the number of times the advertisement is placed is not limited to N3. For example, if N3 = 6 times, the advantages and characteristics of the product refer to the unique strengths and competitive advantages of the product, such as unique design and lower price.

[0130] Specifically, the advertising delivery module generates advertising delivery strategies based on the probability classification of user product purchases, and dynamically adjusts the strategies in conjunction with purchase time intervals and product promotion indices to improve the targeting and flexibility of advertising delivery.

[0131] Specifically, the advertising delivery module compares the purchase time interval TG with the preset purchase time interval TG0, judges the key attributes of the purchase time interval TG based on the comparison results, and adjusts the advertising delivery strategy based on the judgment results, wherein:

[0132] When TG≤TG0, the advertising delivery module determines that the key attribute of the purchase time interval TG is a short time interval and does not adjust the advertising delivery strategy.

[0133] When TG > TG0, the advertising delivery module determines that the key attribute of the purchase time interval TG is a long time interval and adjusts the advertising delivery strategy accordingly.

[0134] When the advertising delivery module adjusts the advertising delivery strategy, for users with low purchase probability, the adjusted number of deliveries is set to N1`, N1` = N1. After a preset delivery duration, the probability G of the user purchasing the product is updated, and the purchase update coefficient is set to β, 1.1 ≤ β ≤ 1.5. The updated probability of the user purchasing the product is G`, G` = β × G.

[0135] For users with a medium probability of purchasing, the adjusted number of deliveries is set to N2`, where N2` = 1.2 × N2;

[0136] For users with a high probability of purchasing, the adjusted number of deliveries is set to N3`, where N3` = 2.5 × N3;

[0137] The advertising module obtains the product's sales growth rate M1, inventory level M2, and competitor product sales growth rate M3 from market research reports. It calculates the product promotion index C using the formula C = 0.4 × M1 + 0.2 × M2 + 0.4 × M3, compares it with a preset promotion index C0, judges the impact of the comparison on the product promotion index C based on the comparison results, and updates the purchase time interval TG based on the judgment results.

[0138] When C≤C0, the advertising delivery module determines that the influence of the product promotion index C is low and does not update the purchase time interval TG.

[0139] When C > C0, the advertising delivery module determines that the influence of the product promotion index C is high and updates the purchase time interval TG. For users with low purchase probability, N1` = N2` = N3` is set; for users with medium purchase probability, N1` = N2` = N3` is set; for users with high purchase probability, no adjustment is made to the advertising delivery strategy.

[0140] Specifically, the preset purchase interval TG0 refers to a pre-set time length threshold used to judge the key attributes of the purchase interval. This embodiment does not limit the preset purchase interval TG0; in this embodiment, TG0 is set to 25 days. The market research report refers to a report written by a professional market research institution after conducting extensive market surveys, collecting data, and analyzing research. This embodiment does not limit the professional market research institution; those skilled in the art can freely choose according to actual needs. The product sales growth rate refers to the percentage increase in the product's sales volume and sales amount within a certain period compared to the sales volume and sales amount of the previous period. This embodiment does not limit the certain period. A certain period is set as one month. The inventory refers to the quantity of products that have not yet been sold. The sales growth rate of competing products refers to the growth rate of the sales quantity and sales amount of competing products within a certain period compared with the sales quantity and sales amount of the previous period. Competing products refer to products of other companies that have a direct competitive relationship with the product. The preset promotion index C0 refers to a pre-set standard value used to measure the degree of influence of the product promotion index. This embodiment does not limit the preset promotion index C0. In this embodiment, C0 = 0.5 is set. The preset placement duration refers to a pre-set time length used to measure the duration of advertising placement. The purchase update coefficient β refers to the coefficient for updating the probability G of users purchasing products.

[0141] Specifically, the advertising delivery module adjusts the probability of users purchasing products by combining the purchase time interval and the promotion index, so as to optimize the delivery frequency for different user segments, thereby improving the accuracy and market adaptability of advertising delivery. For users with a medium purchase probability, the adjusted delivery frequency is set to N2` = 1.2 × N2, so as to maintain user attention while avoiding excessive disturbance. By moderately increasing the reach frequency, the perceived value of the product is strengthened, and users are gradually driven to high-intention conversion. For users with a medium purchase probability, the adjusted delivery frequency is set to N2` = 2.5 × N2, so as to increase the adjusted delivery frequency, thereby seizing the window of potential user demand or leveraging market popularity to accelerate decision-making and increase the purchase probability of users with medium intention.

[0142] Specifically, the monitoring and feedback module obtains the product investment cost I0 and the product investment return I based on market research institution reports, and then uses the formula... The Return on Investment (ROI) for advertising is calculated and compared with a preset ROI of 0. Based on the comparison results, the return attribute of the ROI is determined, and the advertising strategy is optimized accordingly.

[0143] When ROI≥ROI0, the monitoring and feedback module determines that the return attribute of the advertising investment return rate ROI is high and does not optimize the advertising placement strategy.

[0144] When ROI < ROI 0, the monitoring and feedback module determines that the return attribute of the advertising investment ROI is low and optimizes the advertising placement strategy accordingly.

[0145] For users with a low probability of purchase, the first optimization coefficient is set to θ1, where θ1 = 1.7 - 3.2e -1.1×[(ROIO-ROI)-2 The optimized number of deliveries is N1``, N1``=θ1×N1`;

[0146] For users with a medium probability of purchasing, the second optimization coefficient is set to θ2, where θ2 = 1.2 - 3.3e. -1.2×[(ROIO-ROI)+1] The optimized number of deliveries is N2``, N2``=θ2×N2`;

[0147] For users with a high probability of purchase, the third optimization coefficient is set to θ3, where θ3 = 1.5 - 3.2e -1.2×[(ROIO-ROI)+2] The optimized number of deliveries is N3``, N3``=θ3×N3`.

[0148] Specifically, the product investment cost I0 refers to the total cost incurred during the product advertising campaign; the product investment return I refers to the economic benefit obtained from the product advertising campaign; the advertising return on investment (ROI) is an indicator that measures the revenue generated by the advertising campaign based on the product investment cost and product investment return; the preset ROI 0 refers to a pre-set value used to judge the return attribute of the advertising ROI. This embodiment does not limit the advertising ROI; in this embodiment, the advertising ROI is set to 160%. The first optimization coefficient is the coefficient used to optimize the number of ad placements for users with low purchase probability when the advertising ROI is determined to be low; the second optimization coefficient is the coefficient used to optimize the number of ad placements for users with medium purchase probability when the advertising ROI is determined to be low; and the third optimization coefficient is the coefficient used to optimize the number of ad placements for users with high purchase probability when the advertising ROI is determined to be low.

[0149] Specifically, the monitoring and feedback module calculates the return on investment (ROI) of advertising to optimize the advertising strategy, thereby improving the economic efficiency and model performance of advertising. For users with low purchase probability, a first optimization coefficient of θ1 = 0.8 is set to reduce ineffective exposure to low-intent users, lower advertising costs, focus resources on more potential user groups, and avoid over-reaching which could cause user aversion, thus improving the economic efficiency of the advertising strategy. For users with medium purchase probability, a second optimization coefficient of θ2 = 1.5 is set to appropriately increase the frequency of reach to strengthen users' memory of the product value when the ROI is low, thereby balancing cost and conversion efficiency. For users with high purchase probability, a third optimization coefficient of θ3 = 2.0 is set to shorten the decision-making cycle through high-frequency, precise reach to high-intent users, thereby rapidly improving overall advertising revenue.

[0150] Please see Figure 2 As shown, this is a flowchart illustrating the intelligent advertising delivery method based on user behavior analysis in this embodiment. The method includes:

[0151] Step S01 is used to collect target advertising data;

[0152] Step S02 is used to obtain the target comprehensive feature vector based on the target advertising data, to construct the user behavior analysis model using the user behavior analysis model construction method, to construct the cold start model using the cold start model construction method, and to optimize the user behavior analysis model based on the cold start model.

[0153] Step S03 is used to generate an advertising strategy based on a user behavior analysis model using an advertising strategy generation method, and also to adjust the advertising strategy in real time.

[0154] Step S04 is used to calculate the advertising return on investment, obtain the advertising return on investment, and optimize the advertising placement strategy based on the advertising return on investment.

[0155] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. An intelligent advertising delivery system based on user behavior analysis, characterized in that, The system includes: The data acquisition module is used to collect target advertising data; The behavior analysis module is used to obtain the target comprehensive feature vector based on the target advertising data, to build the user behavior analysis model using the user behavior analysis model building method, to build the cold start model using the cold start model building method, and to optimize the user behavior analysis model based on the cold start model. The advertising delivery module is used to generate advertising delivery strategies based on user behavior analysis models using the advertising delivery strategy generation method, and also to adjust the advertising delivery strategies in real time. The monitoring and feedback module is used to calculate the return on advertising investment, obtain the advertising ROI, and optimize the advertising strategy based on the advertising ROI; The advertising delivery module generates an advertising delivery strategy using an advertising delivery strategy generation method, which includes: Step K01: The probability G of a user purchasing a product is obtained through a user behavior analysis model and compared with various preset purchase probabilities, including a first preset purchase probability G1 and a second preset purchase probability G2. Based on the comparison results, the likelihood of the user purchasing the product is judged, and the user's purchase attributes are output based on the judgment results. When G≤G1, the advertising delivery module determines that the probability G of a user purchasing a product is low and outputs the user's purchase attribute as a user with low purchase probability. When G1 < G ≤ G2, the advertising delivery module determines the probability G of a user purchasing a product as medium probability and outputs the user's purchase attribute as a user with medium purchase probability. When G > G2, the advertising delivery module determines that the probability G of a user purchasing a product is high and outputs the user's purchase attribute as a user with a high probability of purchasing. Step K02: When the user's purchase attribute output is "low purchase probability user", the advertising strategy output is: N1 times within a preset period, the advertising content is popular products and newly launched products, providing product information and purchase guidance; When the user's purchase attribute output is "medium purchase probability user", the advertising strategy output will be: N2 times within a preset period, the advertising content will be the value and practicality of the product, providing product information and purchase guidance; When the user's purchase attribute output is "high purchase probability user", the advertising strategy output will be: N3 times within a preset period, and the advertising content will be the advantages and features of the product, providing product information and purchase guidance; The advertising delivery module compares the purchase time interval TG with the preset purchase time interval TG0, judges the key attributes of the purchase time interval TG based on the comparison results, and adjusts the advertising delivery strategy according to the judgment results, wherein: When TG≤TG0, the advertising delivery module determines that the key attribute of the purchase time interval TG is a short time interval and does not adjust the advertising delivery strategy. When TG > TG0, the advertising delivery module determines that the key attribute of the purchase time interval TG is a long time interval and adjusts the advertising delivery strategy accordingly. When the advertising delivery module adjusts the advertising delivery strategy, for users with low purchase probability, the adjusted number of deliveries is set to N1`, N1`=N1. After a preset delivery duration, the probability G of the user purchasing the product is updated, and the purchase update coefficient is set to β, 1.1≤β≤1.

5. The updated probability of the user purchasing the product is G`, G`=β×G. For users with a medium probability of purchasing, the adjusted number of deliveries is set to N2`, where N2` = 1.2 × N2; For users with a high probability of purchasing, the adjusted number of deliveries is set to N3`, where N3` = 2.5 × N3; The advertising module obtains the product's sales growth rate M1, inventory level M2, and competitor product sales growth rate M3 from market research reports. It calculates the product promotion index C using the formula C = 0.4 × M1 + 0.2 × M2 + 0.4 × M3, compares it with a preset promotion index C0, judges the impact of the comparison on the product promotion index C, and updates the purchase time interval TG based on the judgment result. When C≤C0, the advertising delivery module determines that the influence of the product promotion index C is low and does not update the purchase time interval TG. When C > C0, the advertising delivery module determines that the influence of the product promotion index C is high and updates the purchase time interval TG. For users with low purchase probability, N1`=N2`=N3` is set; for users with medium purchase probability, N1`=N2`=N3` is set; for users with high purchase probability, no adjustment is made to the advertising delivery strategy.

2. The intelligent advertising delivery system based on user behavior analysis according to claim 1, characterized in that, The data acquisition module collects target advertising data, which includes e-commerce data and social media data. Specifically, e-commerce data is collected by embedding code in e-commerce websites and mobile applications, and social media data is collected through application programming interfaces (APIs). The e-commerce data includes the number of views, clicks, purchases, the price of purchased goods, the purchase time interval, and the user's browsing time on the platform. The social media data includes user comments, user reposts, user likes, user shares, the number of followers in the user's social circle, and historical advertising records on the e-commerce platform.

3. The intelligent advertising delivery system based on user behavior analysis according to claim 1, characterized in that, The behavior analysis module obtains the target comprehensive feature vector based on the target advertising data, and the process includes: Step J01: Calculate the browsing frequency d1 according to the formula d1=a1 / t, calculate the click frequency d2 according to the formula d2=a2 / t, and calculate the purchase frequency d3 according to the formula d3=a3 / t, where a1 is the number of views, a2 is the number of clicks, a3 is the number of purchases, and t is the preset duration. Use the browsing frequency d1, click frequency d2, and purchase frequency d3 as behavioral frequency features. Step J02: Obtain product association features according to product association rules; Step J03: Compare the price h of the purchased item with the average price h0 of similar items. Based on the comparison result, determine the price sensitivity of the purchased item h, and output the price sensitivity based on the determination result. When h < h0, the price h of the purchased goods is judged to be too high, and the price sensitivity output is low, and low sensitivity is used as the price sensitivity feature. When h≥h0, the price h of the purchased goods is determined to be low, and the price sensitivity output is high, and high sensitivity is used as the price sensitivity feature. Step J04: Use behavioral frequency features, product association features, and price sensitivity features as e-commerce data features; Step J05: Perform sentiment analysis on user comments using natural language processing to obtain sentiment tendency features; perform interest analysis on user reposts, likes, and shares using natural language processing to obtain interest topic features; perform social analysis on user shares and the number of followers in the user's social circle using natural language processing to obtain social influence features; and use sentiment tendency features, interest topic features, and social influence features as social media data features. Step J06: Perform data normalization processing on the e-commerce data features and social media data features to obtain the target comprehensive feature vector.

4. The intelligent advertising delivery system based on user behavior analysis according to claim 3, characterized in that, The behavior analysis module constructs a user behavior analysis model using a user behavior analysis model construction method, which includes: Step U01: Divide the historical comprehensive database into a 70% user behavior training set, a 20% user behavior validation set, and a 10% user behavior test set. Step U02: Set the number of weak learners in the gradient boosting decision tree model parameters to 100 and the learning rate to 0.

1. Train the gradient boosting decision tree model with the user behavior training set input parameters to obtain the trained gradient boosting decision tree model. Step U03: Input the user behavior validation set into the trained gradient boosting decision tree model, optimize the parameters of the trained gradient boosting decision tree model to obtain the parameter-optimized gradient boosting decision tree model, and then input the user behavior test set into the parameter-optimized gradient boosting decision tree model for testing, and output the test accuracy R.

5. The intelligent advertising delivery system based on user behavior analysis according to claim 4, characterized in that, The user behavior analysis model construction method further includes: step U04, comparing the test accuracy R with the preset test accuracy R0, judging the compliance of the parameter-optimized gradient boosting decision tree model based on the comparison result, and outputting the parameter-optimized gradient boosting decision tree model based on the judgment result.

6. The intelligent advertising delivery system based on user behavior analysis according to claim 5, characterized in that, The behavior analysis module constructs a cold start model using a cold start model construction method, which includes: Step Q01 involves comparing the preset ad placements with the historical ad data search database. Based on the comparison results, the existence of the preset ad placements in the historical ad data search database is determined, and the definition of the preset ad placements is output based on the determination result. Specifically: When the preset ad does not exist in the historical ad data search database, the definition of the preset ad will be output as a new ad. When the existence of the preset ad in the historical ad data search database is found to be "existing", the definition of the preset ad will be output as the old ad. Step Q02 involves comparing the preset advertising e-commerce platform with the historical advertising e-commerce platform database. Based on the comparison results, the existence of the preset advertising e-commerce platform in the historical advertising e-commerce platform database is determined, and the definition of the preset advertising e-commerce platform is output based on the determination result. Specifically: When the preset advertising e-commerce platform does not exist in the historical advertising e-commerce platform database, the definition of the preset advertising e-commerce platform will be output as a new platform. When the existence of the preset advertising e-commerce platform in the historical advertising e-commerce platform database is found to be true, the definition of the preset advertising e-commerce platform will be output as the old platform. Step Q03: Compare the user's browsing time t1 within a preset time period with the preset browsing time t0. Based on the comparison result, determine the attributes of the user's browsing time t1 within the preset time period, and output the user type based on the determination result. Step Q04: Divide 70% of the target comprehensive feature vector into a cold start training set and 30% of the target comprehensive feature vector into a cold start validation set; Step Q05: Select the prototype network algorithm as the cold start model, initialize the weights and biases of the cold start model, input the cold start training set into the cold start model, and calculate the output of the cold start model. Step Q06: Calculate the loss function value based on the output and label of the cold start model, calculate the gradient through the backpropagation algorithm, and update the weights and biases of the cold start model. Repeat the forward propagation, loss function calculation and backpropagation process to obtain the trained cold start model. Step Q07: Input the cold start validation set into the trained cold start model for testing, and output the trained cold start model with an accuracy of 90% as the cold start model.

7. The intelligent advertising delivery system based on user behavior analysis according to claim 6, characterized in that, The behavior analysis module obtains the user's platform browsing time t1 within a preset time t2, calculates the user's like frequency according to the formula Fd=f / t2, where f is the number of likes, calculates the user's comment frequency FP according to the formula FP=f2 / t2, where f2 is the number of comments, and calculates the matching degree P between the user group and the platform according to the formula P=ω1×t1+ω2×Fd+ω3×FP, where ω1 is the weight parameter of the user's platform browsing time t1 within the preset time t2, ω2 is the weight parameter of the user's like frequency, and ω3 is the weight parameter of the user's comment frequency. The matching degree P between the user group and the platform is compared with the preset matching degree P0, and the degree of matching between the user group and the platform is judged according to the comparison result. The cold start model is then adjusted according to the judgment result.

8. The intelligent advertising delivery system based on user behavior analysis according to claim 1, characterized in that, The monitoring and feedback module obtains the product investment cost I0 and product investment return I based on market research reports, and then uses the formula... The Return on Investment (ROI) for advertising is calculated and compared with a preset ROI of 0. Based on the comparison results, the return attribute of the ROI is determined, and the advertising strategy is optimized accordingly. When ROI≥ROI0, the monitoring and feedback module determines that the return attribute of the advertising investment return rate ROI is high and does not optimize the advertising placement strategy. When ROI < ROI 0, the monitoring and feedback module determines that the return attribute of the advertising investment ROI is low and optimizes the advertising placement strategy accordingly. For users with a low probability of purchasing, the first optimization coefficient is set to θ1. The optimized number of deliveries is N1``, N1``=θ1×N1`; For users with a medium probability of purchasing, the second optimization coefficient is set to θ2. The optimized number of deliveries is N2, where N2 = θ2 × N2. For users with a high probability of purchase, the third optimization coefficient is set to θ3. The optimized number of deliveries is N3``, N3``=θ3×N3`.

9. A method for applying to an intelligent advertising delivery system based on user behavior analysis as described in any one of claims 1-8, characterized in that, include: Step S01 is used to collect target advertising data; Step S02 is used to obtain the target comprehensive feature vector based on the target advertising data, to construct the user behavior analysis model using the user behavior analysis model construction method, to construct the cold start model using the cold start model construction method, and to optimize the user behavior analysis model based on the cold start model. Step S03 is used to generate an advertising strategy based on a user behavior analysis model using an advertising strategy generation method, and also to adjust the advertising strategy in real time. Step S04 is used to calculate the advertising return on investment, obtain the advertising return on investment, and optimize the advertising placement strategy based on the advertising return on investment.

Images