An AI-based digital marketing effect optimization method, system, and storage medium

Abstract

The present invention discloses a method, system and storage medium for optimizing digital marketing effects based on artificial intelligence, including: statistically analyzing online live broadcast information to obtain real-time product sales data, dividing the product sales data into product sets according to product types, comparing user preferences for the product sets, and extracting preference data that users are interested in; using an inductive sorting algorithm for the preference data to make preference hypotheses for user preference scenarios, and obtaining potential interest points of users; using a decision tree algorithm for the interest points to establish a prediction model for predicting consumer preference patterns, and obtaining the consumption behavior habits of consumers; using a Cox model for the consumption behavior habits to construct a consumption evaluation index model, obtaining a curve showing the change of consumer consumption probability over time, and accurately positioning the relevance of marketing information in the time push dimension, achieving excellent performance and high efficiency of marketing strategies and improving marketing effects.

Description

Technical Field

[0001] The present invention relates to the technical field of digital marketing, and particularly relates to a method, system and storage medium for optimizing digital marketing effects based on artificial intelligence. Background Art

[0002] With the penetration of Internet technology in various fields, online shopping has become a mainstream shopping method. The massive user data brought by online shopping has promoted the development of marketing information delivery towards precision. The diversification of consumer demands has made precision marketing a means for enterprises to gain advantages in the fierce market competition. Precision marketing refers to that enterprises provide appropriate content to appropriate users at appropriate times through appropriate channels by means of diversified marketing methods, that is, including the precision of users, content, channels and timing in four aspects.

[0003] Currently, consumer behavior data is growing exponentially, and consumers expect to obtain a more personalized shopping experience. At the same time, enterprises need to more accurately locate the target customer group and push appropriate marketing information at the right time point to optimize the marketing effect. Currently, the following problems still exist in the method for optimizing digital marketing effects based on artificial intelligence:

[0004] (1) Existing user behaviors and preferences are dynamically changing. Fixed models may not be able to capture these changes in a timely manner, resulting in inaccurate marketing recommendation results. And how to maintain a certain generalization ability while providing personalized services to adapt to new market trends and user changes is an ongoing challenge;

[0005] (2) Existing digital marketing methods based on artificial intelligence need to perform real-time analysis and prediction when facing massive data, consuming a large amount of computing resources. Especially when dealing with large-scale data, this poses a high requirement for the performance of the system. Summary of the Invention

[0006] The purpose of the present invention is to provide a method, system and storage medium for optimizing digital marketing effects based on artificial intelligence to solve the technical problems in the prior art such as inaccurate marketing recommendation results, inability to adapt to new market trends and user changes, and the need to consume a large amount of computing resources.

[0007] To solve the above technical problems, the present invention specifically provides the following technical solutions:

[0008] In the first aspect of the present invention, a method for optimizing digital marketing effects based on artificial intelligence is provided, including the following steps:

[0009] Statistical online live broadcast information to obtain real-time commodity sales data, divide the commodity sales data into commodity sets according to commodity types, compare user preferences for the commodity sets, and extract preference data that users are interested in;

[0010] Apply an inductive sorting algorithm to the preference data to make preference assumptions for user preference scenarios, and obtain potential user interest points;

[0011] Apply a decision tree algorithm to the interest points to establish a prediction model for predicting consumer preference patterns, and obtain consumers' consumption behavior habits;

[0012] Apply the Cox model to the consumption behavior habits to construct a consumption evaluation index model, obtain the curve of the change of consumers' consumption probability over time, and accurately locate the relevance of marketing information in the time push dimension according to the curve to optimize the marketing effect in real time.

[0013] As a preferred embodiment of the present invention, count the online live broadcast information to obtain real-time commodity sales data, and divide the commodity sales data into commodity sets according to commodity types, including:

[0014] Obtain commodity sales information according to real-time live broadcast data, format data in different sources into a unified standard after data screening of the commodity sales information;

[0015] Determine the standard for commodity classification according to business requirements and market research, and automatically classify commodities using a clustering analysis algorithm;

[0016] Conduct trend analysis on the sales data of commodities after automatic classification to obtain the popularity of corresponding commodities, and classify commodities with similar sales characteristics and the same popularity into the same set;

[0017] Regularly evaluate the accuracy of the clustering analysis algorithm, and adjust the classification standard and algorithm parameters according to the actual sales situation to continuously update and optimize the commodity set.

[0018] As a preferred embodiment of the present invention, compare users' preferences for the commodity sets, and extract preference data that users are interested in, including:

[0019] Describe user attribute characteristics with sparse parameters according to the popularity of commodities to obtain a feature vector representing user preferences;

[0020] Mutually multiply the feature vectors and then perform interaction processing to obtain cross features, and perform high-order perception processing on the cross features using an epipolar cross network for heterogeneous features to obtain high-order interaction features;

[0021] Perform feature interaction on the high-order interaction features through the splicing layer and interaction layer of a deep neural network, capture non-linear interaction using MPL open source software, and obtain real-time preference data that users are interested in.

[0022] As a preferred embodiment of the present invention, an inductive sorting algorithm is used for the preference data to make preference hypotheses for user preference scenarios, and potential interest points of users are obtained, including:

[0023] The preference data is inductively sorted according to the commodity set of the corresponding consumer goods by using the MPR minimum privilege criterion. By extracting the differences in the preference data to search for the corresponding commodities, the corresponding commodities are used as the target columns;

[0024] The target columns are mined by using the MAP list-level index to obtain the preference differences between commodities and incorporate them into the pairwise preference;

[0025] The pairwise preference is evaluated and quantified by using a pairwise objective function, a multi-objective pairwise sorting framework is constructed, and the Pareto optimal solution of the pairwise sorting is obtained , and its expression is:

[0026] ;

[0027] Among them, represents the parameter set of the scoring model in the pairwise objective function, , are respectively , 's two tentative losses;

[0028] Based on the Pareto optimal solution optimize the sensitive quality of the ranking list to obtain potential interest points of users.

[0029] As a preferred embodiment of the present invention, a decision tree algorithm is used for the interest points to establish a prediction model for predicting consumer preference patterns; consumer consumption behavior habits are obtained, including:

[0030] Build a decision tree model through python, and set the attribute screening criterion as the information gain rate to obtain the most discriminative features;

[0031] Set the feature division points corresponding to the features to a random mode, set the feature selection method to an entropy-based mode, and adjust the class weights according to the proportion of four types of samples;

[0032] Randomly select 70% of the interest points as the training data set, and the remaining 30% of the interest points as the test data set for cross-validation to obtain the maximum tree depth and the maximum number of leaf nodes;

[0033] Deploy the best decision tree model to make real-time predictions on new consumer behavior data and obtain the key factors affecting their online shopping decisions;

[0034] Based on the output result of the decision tree model, obtain the real-time consumption behavior habits of consumers.

[0035] As a preferred embodiment of the present invention, a consumption evaluation index model is constructed for the consumption behavior habits by using the Cox model, and a curve showing the change of the consumption probability of consumers over time is obtained, including:

[0036] Use a multi-layer perceptron MLP for the consumption behavior habits to capture the cross-behavior habits of consumers, and extract the factors affecting consumer behavior as covariates, including the age, gender, income level, and shopping frequency of consumers;

[0037] Define the time variable and the probability of event occurrence. After removing missing values, outliers, and duplicate data, use the lifelines library in Python to construct the Cox model;

[0038] Evaluate the goodness of fit of the model by calculating the log-likelihood ratio of the Cox model, and obtain the curve of the change of the consumption probability over time.

[0039] As a preferred embodiment of the present invention, the Cox model consists of a hazard function and a survival function, and the expression of the hazard function is:

[0040] ;

[0041] where represents the covariate, represents the regression coefficient of the linear equation, is the baseline hazard function, indicating the probability that the research object has a specific event at time t without being affected by the covariate X;

[0042] The expression of the survival function is:

[0043] ;

[0044] ;

[0045] where represents the baseline survival function of the research object at time t.

[0046] As a preferred embodiment of the present invention, accurately locate the relevance of marketing information in the time push dimension according to the curve, and optimize the marketing effect in real time, including:

[0047] Extract the time period with a high probability of consumer purchase behavior from the curve of the change of the consumption probability over time, and use this time period as the best marketing time period;

[0048] According to the best marketing time period of each consumer, formulate a personalized marketing information push plan, determine the type of push content and customize it in combination with the historical preferences of consumers;

[0049] Process the customized information through the database in the form of a message queue, integrate the real-time data stream, and update the consumption probability curve of consumers in real time;

[0050] According to the consumption probability curve, dynamically adjust the push time and frequency of marketing information; build a marketing automation platform to achieve automated marketing information push, set trigger conditions, and automatically send marketing information.

[0051] The second aspect of the present invention provides a system for optimizing the digital marketing effect based on artificial intelligence, including:

[0052] Data processing module: Collect consumer behavior data and live broadcast data in real time, store massive data using a distributed database and a big data storage system, and perform real-time processing on the massive data to obtain a data training set;

[0053] Marketing strategy prediction module: Train the data training set using the decision tree algorithm, and use the Cox model to predict the change of consumers' purchase probability over time;

[0054] Marketing effect optimization module: Generate a personalized product recommendation list based on the user's historical behavior and real-time data, and formulate and execute a personalized marketing information push plan according to the consumer's consumption probability curve;

[0055] Real-time monitoring and feedback module: Real-time monitor the effect of marketing activities, collect click-through rate and conversion rate indicators, automatically adjust the push strategy, and continuously optimize the marketing effect.

[0056] The present invention has the following beneficial effects compared with the prior art:

[0057] The present invention adopts a multi-pair ranking algorithm. By dividing products into different product sets and comparing the preference differences of users for these product sets, it can better explore the potential interests of users. It uses the Cox model to predict the change of consumers' purchase probability over time, and based on the user's historical behavior and real-time data, generates a personalized product recommendation list. According to the consumer's consumption probability curve, it formulates and executes a personalized marketing information push plan, real-time monitors the effect of marketing activities, collects key indicators such as click-through rate and conversion rate, automatically adjusts the push strategy, and continuously optimizes the marketing effect, achieving superior performance and high efficiency of marketing strategies in various scenarios and improving the marketing effect. Brief Description of the Drawings

[0058] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are merely exemplary. For those of ordinary skill in the art, without creative efforts, other implementation drawings can also be obtained based on the provided drawings.

[0059] Figure 1 Flowchart of the digital marketing effect optimization method based on artificial intelligence provided by the embodiment of the present invention;

[0060] Figure 2 Block diagram of the digital marketing effect optimization system based on artificial intelligence provided by the embodiment of the present invention. Specific embodiments

[0061] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0062] As Figure 1 shown, the present invention provides a digital marketing effect optimization method based on artificial intelligence, including the following steps:

[0063] Statistical online live broadcast information to obtain real-time product sales data, divide the product sales data into product sets according to product types, compare user preferences for the product sets, and extract preference data that users are interested in;

[0064] In this embodiment, by docking with major online live broadcast platforms through APIs, real-time live broadcast data is obtained, which may include the number of viewers in the live broadcast room, interaction situations, the number of product displays, and the sales data directly generated in the live broadcast room. Real-time product sales data is obtained from e-commerce platforms, especially the sales volume and sales amount generated through live broadcast promotion. Duplicate or incomplete data records are removed to ensure the accuracy and effectiveness of the analyzed data.

[0065] Use the inductive sorting algorithm to make preference assumptions for user preference scenarios for the preference data, and obtain potential interest points of users;

[0066] In this embodiment, the inductive sorting algorithm is used to perform preference data induction for user preference scenarios, perform trend analysis on different types of sales data, and obtain potential interest points of users to provide strong support for subsequent digital marketing strategies.

[0067] A prediction model for predicting consumers' preference patterns is established for the said points of interest using a decision tree algorithm to obtain consumers' consumption behavior habits;

[0068] A consumption evaluation index model is constructed for the said consumption behavior habits using a Cox model to obtain the curve of consumers' consumption probability changing over time, and the relevance of marketing information in the time push dimension is accurately positioned according to the curve to optimize the marketing effect in real time.

[0069] In this embodiment, new consumers' behavior data is predicted in real time through the curve of consumers' consumption probability changing over time, the key factors affecting their online shopping decisions are identified, and the factors having a significant impact on online shopping decisions are extracted, such as consumers' shopping experience, attitudes towards financial risks, and attention to privacy risks. These factors can help enterprises better understand consumers' needs and optimize marketing strategies and service experiences.

[0070] Online live broadcast information is statistically analyzed to obtain real-time commodity sales data, and the said commodity sales data is divided into commodity sets according to commodity types, including:

[0071] Commodity sales information is obtained according to real-time live broadcast data, and after data screening of the said commodity sales information, data from different sources is formatted into a unified standard;

[0072] According to business requirements and market research, the criteria for commodity classification are determined, and the clustering analysis algorithm is used to automatically classify commodities;

[0073] Trend analysis is performed on the sales data of the automatically classified commodities to obtain the popularity of the corresponding commodities, and the commodities with similar sales characteristics and the same popularity are grouped into the same set;

[0074] The accuracy of the clustering analysis algorithm is regularly evaluated, and the classification criteria and algorithm parameters are adjusted according to the actual sales situation to continuously update and optimize the commodity sets.

[0075] In this embodiment, by regularly evaluating the accuracy of the clustering analysis algorithm, the commodities with similar sales characteristics are grouped into the same set. These commodity sets can be used for further market analysis, marketing strategy formulation, and inventory management, etc., and the classification criteria and algorithm parameters are adjusted according to the actual sales situation. Based on the new sales data and market changes, the commodity sets are continuously updated and optimized to adapt to the dynamic changes of the market.

[0076] The preferences of users for the said commodity sets are compared, and the preference data of interest to users is extracted, including:

[0077] The user attribute characteristics are described with sparse parameters according to the popularity of commodities to obtain the feature vector representing user preferences;

[0078] In this embodiment, the user attribute features are represented by sparse parameters, such as "gender=female, emotional status=single", etc. Each of these features is formalized into a feature vector related to the classification field, and these feature vectors are called sparse features.

[0079] Multiply the feature vectors with each other and then perform interaction processing to obtain cross features. Perform high-order perception processing on the cross features using an epipolar cross network for heterogeneous features to obtain high-order interaction features;

[0080] Perform feature interaction on the high-order interaction features through the splicing layer and interaction layer of the deep neural network, capture the non-linear interaction using the MPL open-source software, and obtain the real-time preference data of the user's interest.

[0081] In this embodiment, in the feature vector interaction, more attention was paid to the product of sparse features before the cross features, showing the prospect of improving the prediction performance.

[0082] In this embodiment, an epipolar cross network for heterogeneous features is adopted to construct cross features oriented to the feature structure. Using XCrossNet for modeling includes three stages. In the feature construction stage, a cross layer is proposed to construct dense cross features and an inner product layer is proposed to construct sparse cross features. In the feature splicing stage, the dense cross features and sparse cross features interact through a splicing layer and a cross layer. Finally, in the feature selection stage, the MLP is used to capture the non-linear interaction and its relative importance.

[0083] Perform preference hypothesis on the preference data using an inductive ranking algorithm for the user preference scenario to obtain the potential interest points of the user, including:

[0084] Sort the preference data according to the commodity set of the corresponding consumer goods using the MPR minimum privilege criterion, search for the corresponding commodities by extracting the differences in the preference data, and use the corresponding commodities as the target columns;

[0085] Mine the target columns using the MAP list-level metric to obtain the preference differences between commodities and incorporate them into the pairwise preference;

[0086] Evaluate and quantify the pairwise preference using a pairwise objective function, construct a multi-objective pairwise ranking framework, and obtain the Pareto optimal solution of the pairwise ranking , and its expression is:

[0087] ;

[0088] Among them, represents the parameter set of the scoring model in the pairwise objective function, , respectively and two tentative losses;

[0089] According to the Pareto optimal solution Optimize the sensitive quality of the ranking list to obtain potential user interest points.

[0090] In this embodiment, a multi-objective optimizer based on multi-gradient descent is used to sort the pair-level objective function to optimize the matrix factorization model, so as to find the Pareto optimal solution of the two objective functions. Moreover, the multi-objective pair-level sorting algorithm is applicable to any pair-level sorting algorithm without introducing additional hyperparameters or auxiliary information. The multi-objective pair-level sorting algorithm can not only retain the unique advantages of a single objective in various complex scenarios, but also has high operating efficiency and can more effectively meet the requirements of complex scenarios.

[0091] A prediction model for predicting consumer preference patterns is established for the interest points by using the decision tree algorithm; obtain the consumption behavior habits of consumers, including:

[0092] Build a decision tree model through python, and set the attribute screening criterion as the information gain rate to obtain the most discriminative features;

[0093] In this embodiment, the information gain rate can indirectly eliminate the influence brought by the diversity of feature values. At the same time, in the process of classifying data, it will not interfere with the subsequent classification prediction results due to the current classification results.

[0094] Set the feature division points corresponding to the features in a random mode, set the feature selection method in an entropy-based mode, and adjust the class weights according to the proportion of four types of samples;

[0095] Randomly select 70% of the interest points as the training data set, and the remaining 30% of the interest points as the test data set for cross-validation to obtain the maximum tree depth and the maximum number of leaf nodes;

[0096] Deploy the best decision tree model to make real-time predictions on new consumer behavior data, and obtain the key factors affecting their online shopping decisions;

[0097] Based on the output results of the decision tree model, obtain the real-time consumption behavior habits of consumers.

[0098] In this embodiment, factors that have a significant impact on online shopping decisions are extracted from the decision tree model, such as consumers' shopping experience, attitudes towards financial risks, and attention to privacy risks. These factors can help enterprises better understand consumer needs and optimize marketing strategies and service experiences.

[0099] Construct a consumption evaluation index model for the consumption behavior habits using the Cox model, and obtain the curve of the consumer's consumption probability changing over time, including:

[0100] Use a multi-layer perceptron MLP for the consumption behavior habits to capture the cross-behavior habits of consumers, and extract the factors affecting consumer behavior as covariates, including the age, gender, income level, and shopping frequency of consumers;

[0101] Define the time variable and the probability of event occurrence. After removing missing values, outliers, and duplicate data, use the lifelines library in Python to construct the Cox model;

[0102] Evaluate the goodness of fit of the model by calculating the log-likelihood ratio of the Cox model, and obtain the curve of the consumption probability changing over time.

[0103] In this embodiment, the Cox model is used to construct a consumption evaluation index model and obtain the curve of the consumer's consumption probability changing over time. This method can help enterprises better understand the consumer behavior pattern, formulate more effective marketing strategies, flexibly adjust marketing strategies and service experiences, and improve customer satisfaction and conversion rate.

[0104] The Cox model consists of a hazard function and a survival function. The expression of the hazard function is:

[0105] ;

[0106] Among them, represents the covariate, represents the regression coefficient of the linear equation, is the baseline hazard function, indicating the probability of the research object having a specific event at time t when not affected by the covariate X;

[0107] In this embodiment, all covariates in the Cox model need to satisfy the proportional hazards assumption, that is, the value of the baseline survival function is only related to time t and is not affected by the value of the covariate . Therefore, before establishing the Cox model, it is necessary to first test the proportional hazards assumption of the variables.

[0108] The expression of the survival function is:

[0109] ; [[ID=4N]]

[0110] ;

[0111] Among them, represents the baseline survival function of the research object at time t.

[0112] In this embodiment, the graph of the survival function is called the survival curve. The changing trend of the survival curve can reflect the change of the individual survival probability over time. A smooth survival curve represents a longer survival time or a higher survival probability; conversely, a steep curve represents a shorter survival time or a lower survival probability.

[0113] In this embodiment, viewing historical order footprints, chatting with customer service, adding items to the shopping cart, and receiving shopping vouchers are all risk factors. That is, after consumers have the above behaviors, the probability of being active in this stage will decrease. From a practical perspective, viewing historical order footprints and chatting with customer service both mean that consumers have formed a purchase plan and need to rely on their own purchase experience and customer service to help them make decisions. Once they obtain the corresponding information, they will move to the next stage. While browsing the shopping cart is a protective factor, that is, consumers are more likely to stay in the current stage after browsing the shopping cart. This is because browsing the shopping cart gives consumers a chance to discover more shopping plans and evaluate them.

[0114] Precisely locate the relevance of marketing information in the time push dimension according to the curve, and optimize the marketing effect in real time, including:

[0115] Extract the time period with a high probability of consumer purchase behavior from the curve of the change of the consumption probability over time, and use this time period as the best marketing time period;

[0116] According to the best marketing time period of each consumer, formulate a personalized marketing information push plan, determine the type of push content and customize it in combination with the historical preferences of consumers;

[0117] Process the customized information in the form of a message queue through the database, integrate the real-time data stream, and update the consumption probability curve of consumers in real time;

[0118] According to the consumption probability curve, dynamically adjust the push time and frequency of marketing information; build a marketing automation platform to achieve automated marketing information push, set trigger conditions, and automatically send marketing information.

[0119] In this embodiment, based on the curve of the change of the consumption probability of consumers over time, precisely locate the relevance of marketing information in the time push dimension, and optimize the marketing effect in real time, which can not only improve the pertinence and efficiency of marketing activities, but also enhance the customer experience and satisfaction.

[0120] Second Embodiment: A system for an artificial intelligence-based digital marketing effect optimization method, including:

[0121] Data processing module: Real-time collect consumer behavior data and live broadcast data, store massive data using a distributed database and a big data storage system, and perform real-time processing on the massive data to obtain a data training set;

[0122] Marketing strategy prediction module: Use the decision tree algorithm to train the data training set, and use the Cox model to predict the change of consumer purchase probability over time;

[0123] Marketing effect optimization module: Generate a personalized product recommendation list based on the user's historical behavior and real-time data, and formulate and execute a personalized marketing information push plan according to the consumer's consumption probability curve;

[0124] Real-time monitoring and feedback module: Real-time monitor the effect of marketing activities, collect click-through rate and conversion rate indicators, automatically adjust the push strategy, and continuously optimize the marketing effect.

[0125] In this embodiment, starting from consumer behavior data, real-time live broadcast data, and external data, key features are extracted, such as the user's shopping frequency, preference category, financial risk tolerance, etc. The data is divided into a training set, a validation set, and a test set, and the Cox model is used to predict the change of consumer purchase probability over time. Based on the user's historical behavior and real-time data, a personalized product recommendation list is generated, and a personalized marketing information push plan is formulated and executed according to the consumer's consumption probability curve. The effect of marketing activities is real-time monitored, key indicators such as click-through rate and conversion rate are collected, the push strategy is automatically adjusted, and the marketing effect is continuously optimized.

[0126] Third embodiment: A storage medium,

[0127] The storage medium stores computer execution instructions, and when the processor executes the computer execution instructions, the artificial intelligence-based digital marketing effect optimization method of this embodiment is implemented.

[0128] The present invention adopts a multi-pair ranking algorithm. By dividing products into different product sets and comparing the preference differences of users for these product sets, the potential interests of users can be better explored. The Cox model is used to predict the change of consumer purchase probability over time. Based on the user's historical behavior and real-time data, a personalized product recommendation list is generated, and a personalized marketing information push plan is formulated and executed according to the consumer's consumption probability curve. The effect of marketing activities is real-time monitored, key indicators such as click-through rate and conversion rate are collected, the push strategy is automatically adjusted, and the marketing effect is continuously optimized, achieving superior performance and high efficiency of marketing strategies in various scenarios and improving the marketing effect.

[0129] The above embodiments are only exemplary embodiments of the present application and are not used to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art can make various modifications or equivalent substitutions within the essence and protection scope of the present application, and such modifications or equivalent substitutions should also be regarded as falling within the protection scope of the present application.

Claims

1. A method for optimizing the effect of digital marketing based on artificial intelligence, characterized in that, It includes the following steps: Statistically analyze the online live broadcast information to obtain real-time product sales data, divide the product sales data into product sets according to product types, compare the preferences of users for the product sets, and extract the preference data that users are interested in; Use the inductive sorting algorithm for the preference data to make preference hypotheses for user preference scenarios, and obtain potential interest points of users; Use the decision tree algorithm for the interest points to establish a prediction model for predicting consumer preference patterns, and obtain the consumption behavior habits of consumers; Use the Cox model for the consumption behavior habits to construct a consumption evaluation index model, obtain the curve of the change of the consumption probability of consumers over time, accurately locate the relevance of marketing information in the time push dimension according to the curve, and optimize the marketing effect in real time; Use the inductive sorting algorithm for the preference data to make preference hypotheses for user preference scenarios, and obtain potential interest points of users, including: Summarize and organize the preference data according to the product sets of corresponding consumer products using the MPR minimum privilege criterion, search for corresponding products by extracting the differences in the preference data, and use the corresponding products as the target columns; Mine the target columns using the MAP list-level index, obtain the preference differences between products and incorporate them into the pairwise preference; The pair-level preference is evaluated and quantified using a pair-level objective function, a multi-objective pair-level ranking framework is constructed, and the Pareto optimal solution of the pair-level ranking is obtained , and its expression is: ； Among them, represents the parameter set of the scoring model in the pairwise objective function, , respectively are , two tentative losses of Based on the Pareto optimal solution Optimize the sensitive quality of the ranking list to obtain potential user interest points; Use the decision tree algorithm for the interest points to establish a prediction model for predicting consumer preference patterns; obtain the consumption behavior habits of consumers, including: Build a decision tree model through python, set the attribute screening criterion as the information gain rate to obtain the most discriminative features; Set the feature division points corresponding to the features as the random mode, set the feature selection method as the entropy-based mode, and adjust the class weights according to the proportion of four types of samples; Randomly select 70% of the interest points as the training data set, and use the remaining 30% of the interest points as the test data set for cross-validation to obtain the maximum tree depth and the maximum number of leaf nodes; Deploy the best decision tree model, perform real-time prediction on new consumer behavior data, and obtain the key factors affecting their online shopping decisions; Based on the output results of the decision tree model, obtain the real-time consumption behavior habits of consumers.

2. The method for optimizing digital marketing effect based on artificial intelligence according to claim 1, characterized in that Statistically analyze the online live broadcast information to obtain real-time product sales data, divide the product sales data into product sets according to product types, including: Obtain product sales information according to real-time live broadcast data, format the data in different sources into a unified standard after data screening; Determine the criteria for product classification according to business requirements and market research, and use the clustering analysis algorithm to automatically classify products; Conduct trend analysis on the sales data of automatically classified products to obtain the popularity of corresponding products, and classify products with similar sales characteristics and the same popularity into the same set; Regularly evaluate the accuracy of the clustering analysis algorithm, adjust the classification criteria and algorithm parameters according to the actual sales situation, and continuously update and optimize the product set.

3. The optimization method for digital marketing effect based on artificial intelligence according to claim 2, characterized in that Compare the preferences of users for the commodity set, and extract the preference data that users are interested in, including: Describe the user attribute features with sparse parameters according to the popularity of the commodities, and obtain the feature vector representing the user preferences; After multiplying the feature vectors with each other, perform interactive processing to obtain cross features, and perform high-order perception processing on the cross features using an epipolar cross network for heterogeneous features to obtain high-order interactive features; Perform feature interaction on the high-order interactive features through the splicing layer and interactive layer of the deep neural network, and capture the non-linear interaction using the MPL open-source software to obtain the real-time preference data that users are interested in.

4. The optimization method for digital marketing effect based on artificial intelligence according to claim 1, characterized in that Construct a consumption evaluation index model for the consumption behavior habits using the Cox model, and obtain the curve of the change of the consumption probability of consumers over time, including: Use a multi-layer perceptron MLP to capture the cross-behavior habits of consumers for the consumption behavior habits, and extract the factors affecting consumer behavior as covariates, including the age, gender, income level and shopping frequency of consumers; Define the time variable and the probability of event occurrence, and after removing missing values, outliers and duplicate data, use the lifelines library in Python to construct the Cox model; Evaluate the goodness of fit of the model by calculating the log-likelihood ratio of the Cox model, and obtain the curve of the change of the consumption probability over time.

5. The method for optimizing digital marketing effect based on artificial intelligence according to claim 4, characterized in that, Including: The Cox model consists of a hazard function and a survival function, and the expression of the hazard function is: ； Among them, represents the covariate, represents the regression coefficient of the linear equation, is the baseline hazard function, representing the probability that the subject will experience a specific event at time t when not affected by the covariate X; The expression of the survival function is: ； ； Among them, represents the benchmark survival function of the research object at time t.

6. The optimization method for digital marketing effect based on artificial intelligence according to claim 5, characterized in that Accurately locate the relevance of marketing information in the time push dimension according to the curve, and optimize the marketing effect in real time, including: Extract the time period with a high probability of consumer purchase behavior from the curve of the change of the consumption probability over time, and use this time period as the best marketing time period; According to the best marketing time period of each consumer, formulate a personalized marketing information push plan, determine the type of push content and customize it in combination with the historical preferences of consumers; Process the customized information in the form of a message queue through the database, integrate the real-time data stream, and update the consumption probability curve of consumers in real time; According to the consumption probability curve, dynamically adjust the push time and frequency of marketing information; build a marketing automation platform to realize automated marketing information push, set trigger conditions, and automatically send marketing information.

7. A system for an optimization method of digital marketing effect based on artificial intelligence, which is used to implement the method described in any one of claims 1-6, characterized in that, Including: Data processing module: Real-time collect consumer behavior data and live broadcast data, use a distributed database and a big data storage system to store massive data, and perform real-time processing on the massive data to obtain a data training set; Marketing strategy prediction module: Use the decision tree algorithm to train the data training set, and use the Cox model to predict the change of the consumer purchase probability over time; Marketing Effect Optimization Module: Generate a personalized product recommendation list based on users' historical behaviors and real-time data, and formulate and execute a personalized marketing information push plan according to the consumption probability curve of consumers; Real-time Monitoring and Feedback Module: Monitor the effect of marketing activities in real time, collect click-through rate and conversion rate indicators, automatically adjust the push strategy, and continuously optimize the marketing effect.

8. A storage medium, characterized in that, Including: The computer-executable instructions are stored in the storage medium, and when the processor executes the computer-executable instructions, the method described in any one of claims 1-6 is implemented.

Images