Advertisement marketing system based on data updating and sorting

Abstract

The invention relates to the technical field of advertisement data marketing, in particular to an advertisement marketing system based on data updating and sorting, which comprises a data acquisition and transmission unit, a data processing and analysis unit, an advertisement strategy generation and optimization unit and an advertisement putting monitoring unit. The data processing and analyzing unit is used for calculating user behaviors and market trend indexes, mining atlas association and predicting the user behaviors by applying memory calculation, a parallel processing framework and various algorithms, and the advertisement strategy generating and optimizing unit is used for constructing a strategy generating engine to generate an initial strategy and dynamically optimizing by utilizing deep reinforcement learning. And the advertisement putting monitoring unit applies the optimized strategy to actual putting and monitors the conversion rate in real time, and detects abnormity and adjusts the strategy through an auto-encoder model, so that the accuracy and efficiency of advertisement marketing are improved, the input-output ratio of advertisement putting is improved, and the advertisement effect is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of advertising data marketing, and in particular to an advertising marketing system based on data updating and organization. Background Art

[0002] Advertising data marketing is an important technology. With the rapid development of the Internet, the field of advertising marketing has undergone tremendous changes, and data-driven advertising marketing strategies have gradually become mainstream.

[0003] Traditional systems do not collect user behavior data, market competition data, and industry trend data comprehensively, and the data cleaning and processing methods are relatively simple, making it difficult to cope with complex and changing data environments, resulting in low data quality and an inability to accurately reflect the true market situation. In addition, they lack efficient processing architectures and advanced algorithms, and are unable to dynamically adjust analysis windows based on real-time data traffic. They are also difficult to quickly and accurately calculate user behavior indicators and market trend indicators, and are also difficult to deeply explore potential correlations between user behavior graphs and market relationship graphs. They are unable to accurately predict future user behavior trends, resulting in the lack of scientificity and accuracy in the generated initial advertising delivery strategies, which are mostly based on experience or simple data analysis and are difficult to adapt to complex market environments. Moreover, strategy adjustments are often delayed and cannot be optimized in real time based on advertising delivery results, resulting in an unsatisfactory input-output ratio for advertising delivery. In order to solve this technical problem, we provide an advertising marketing system based on data updates and collation. Summary of the invention

[0004] The purpose of the present invention is to provide an advertising and marketing system based on data updating and organization to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above purpose, an advertising marketing system based on data updating and sorting is provided, including a data collection and transmission unit, a data processing and analysis unit, an advertising strategy generation and optimization unit, and an advertising delivery monitoring unit;

[0006] The data acquisition and transmission unit acquires user behavior data, market competition data, and industry trend data through a data collector, and cleans and processes the acquired data to obtain a comprehensive data set;

[0007] The data processing and analysis unit uses a data processing module based on memory computing and parallel processing architecture to process the data in the comprehensive data set, and uses an adaptive window algorithm to adjust the window size according to the real-time flow of the data to obtain the real-time change characteristics of the data. At the same time, it combines asynchronous I / O technology and vectorized computing methods to perform real-time aggregation analysis on the data, calculates user behavior indicators and market trend indicators, and stores the user behavior indicators and market trend indicators in a graph database. The data analysis module is then used to extract user behavior graphs and market relationship graphs from the graph database, and the user behavior graphs and market relationship graphs are associated and mined through a graph neural network algorithm based on an attention mechanism to obtain user behavior sequences, and then a long short-term memory network is used to model the user behavior sequences to predict future behavior trends.

[0008] The advertising strategy generation and optimization unit constructs a strategy generation engine through a strategy generation module, and generates an initial advertising delivery strategy in combination with real-time data analysis results and preset business rules. The dynamic optimization algorithm based on multi-agent deep reinforcement learning is introduced through the strategy optimization module, and the strategy parameters are adjusted in real time according to the advertising delivery effect by combining experience replay and priority experience replay. At the same time, a real-time advertising effect evaluation index system is established, and the real-time calculation function of the effect evaluation module is used to conduct real-time evaluation of the advertising input-output ratio;

[0009] The advertising delivery monitoring unit applies the optimized advertising strategy to actual delivery and monitors the conversion rate of the advertisement in real time. It uses an anomaly detection algorithm based on streaming time series analysis to monitor abnormal situations in the advertising delivery process, analyzes the monitoring data in real time, and adjusts the advertising delivery strategy according to the analysis results.

[0010] As a further improvement of the technical solution, the data collection and transmission unit adopts a density-based local outlier factor algorithm for outlier detection in user behavior data during data cleaning. The specific steps are as follows:

[0011] For each data point in the user behavior dataset , calculate it to The distance between the nearest neighbors is is a pre-set parameter, denoted as , and take the first The distance between neighboring points is No. distance ;

[0012] calculate of Distance from all points in the neighborhood to The average distance is denoted as The reachable distance is based on The reachable distance is calculated The local reachable density of Locally reachable density calculation The local outlier factor of If the local outlier factor of is greater than 1, then Determined as an outlier.

[0013] As a further improvement of the technical solution, in the data processing and analysis unit, when the data processing module based on memory computing and parallel processing architecture adopts the adaptive window algorithm, the window size adjustment formula is as follows:

[0014] Set the current window size to , the real-time data flow is The flow rate change rate is ,in Define an adaptive coefficient for the flow at the previous moment ;

[0015] when When the window is resized to ;when When the window is resized to .

[0016] As a further improvement of the technical solution, in the data processing module, when the asynchronous I / O technology and the vectorized calculation method are combined to perform real-time aggregation analysis, the user behavior index is calculated as follows:

[0017] Extracting a user behavior data set from the comprehensive data set of the data acquisition and transmission unit, wherein the user behavior data set includes a user ID, behavior time, and behavior type, and reading data in parallel using asynchronous I / O technology to convert the read data into a vector form;

[0018] The behavior types include browsing, clicking and purchasing. For the behavior types, one-hot encoding is used for vectorization, and the user behavior frequency index is calculated. For each user, the cumulative sum of the different behavior type vectors within a certain time window is counted, and then divided by the length of the time window to obtain the behavior frequency vector, and the user behavior index is calculated based on the behavior frequency vector.

[0019] As a further improvement of the technical solution, in the data analysis module, when the graph neural network algorithm based on the attention mechanism performs association mining on the user behavior graph and the market relationship graph, the specific steps are as follows:

[0020] Record and set the node feature vectors of the user behavior graph and the market relationship graph respectively. For each node in the user behavior graph, obtain the query vector through linear transformation. For each node in the market relationship graph, obtain the key vector and value vector.

[0021] The attention score is calculated based on the query vector and the key vector, and the attention score is combined with the preset softmax function to calculate the attention weight. The associated feature vector is then calculated based on the attention weight and the value vector. The associated feature vector is concatenated with the original node feature vector and then subjected to a nonlinear transformation to obtain an updated node feature vector. The association relationship between the two graphs is mined based on the node feature vector.

[0022] As a further improvement of the technical solution, in the data analysis module, when the long short-term memory network models the user behavior sequence, the gated recurrent unit structure is optimized, and the specific operations are as follows:

[0023] For the input user behavior sequence and the previous hidden state , calculate the reset gate , update gate ,in is the sigmoid function, is a weight matrix, the reset gate is used to control the reset of information in the hidden state of the previous moment, and the update gate is used to control the transmission and update of information;

[0024] The candidate hidden state is calculated based on the user behavior sequence, the hidden state at the previous moment, and the reset gate, and the final hidden state is calculated based on the candidate hidden state. Finally, the long-term dependency in the user behavior sequence is obtained and modeled based on the final hidden state.

[0025] As a further improvement of the present technical solution, in the advertising strategy generation optimization unit, when the strategy generation module constructs the strategy generation engine, a genetic algorithm is used to optimize the generation of the initial advertising delivery strategy. The specific steps are as follows:

[0026] Encode the advertising time, delivery channel and delivery budget into an integer sequence as a chromosome, and randomly generate a certain number of chromosomes as the initial population. Each chromosome represents an initial advertising delivery strategy, and then calculate the fitness based on the expected conversion rate of the advertisement.

[0027] The parent chromosome is selected according to the fitness value, and the parent chromosome is crossover and mutation operations are performed to generate the child chromosome. The above steps are repeated until the maximum number of iterations is met to obtain the optimized initial advertising delivery strategy.

[0028] As a further improvement of the technical solution, the strategy optimization module uses a method combining experience playback and priority experience playback to adjust the strategy parameters in real time according to the advertising delivery effect. The specific operations are as follows:

[0029] During the advertising process, multiple agents interact with the advertising environment. Each agent goes through a state at each time step and selects an action based on the current strategy to perform advertising operations. It then receives a reward, which is set by the advertising conversion rate. It then enters the next state and stores this set of information as a sample in the experience replay pool, which is a first-in, first-out queue. When the queue is full, the earliest sample will be removed.

[0030] A priority is calculated for each sample in the experience replay pool, where the priority is determined by the reward size, timestamp, and prediction error. Each time a batch of samples is sampled from the experience replay pool for training the deep reinforcement learning model, samples are sampled according to the priority of the samples to obtain a batch of samples, which are input into the deep reinforcement learning model for training. According to the deep reinforcement learning algorithm, the loss function is calculated and the parameters of the model are updated through the back propagation algorithm.

[0031] As a further improvement of the present technical solution, in the real-time advertising effect evaluation index system established by the effect evaluation module, when calculating the advertising input-output ratio index, the delay effect of advertising delivery is taken into account. The specific formula is as follows:

[0032] Set the ad delivery time to , after the ad is placed Statistical income within time , the advertising cost is , considering that the revenue will be generated after a period of time after the advertisement is released, the delayed revenue is calculated by weighted average, that is:

[0033] ;

[0034] in yes The profit of the moment, It is based on time and ad delivery time The weight determined by the distance is .

[0035] As a further improvement of the technical solution, in the advertising delivery monitoring unit, the anomaly detection algorithm based on streaming time series analysis adopts an autoencoder model, and the specific steps are as follows:

[0036] Normalizing the monitoring data during the advertising delivery process and using it as the input of the autoencoder, wherein the monitoring data is the advertising conversion rate, and the autoencoder is composed of an encoder and a decoder;

[0037] The encoder maps the input to a low-dimensional representation, and the decoder reconstructs the low-dimensional representation into a new input for the autoencoder. The reconstruction error is defined by calculating the difference between the new input and the input of the autoencoder, and the autoencoder is trained to minimize the reconstruction error.

[0038] During actual detection, for new monitoring data, the reconstruction error is calculated. If the reconstruction error is greater than the preset threshold, the data point is determined to be an abnormal point and real-time monitoring is performed. Based on the abnormal detection results, the abnormal situation is input as a state into the reinforcement learning model, and the model outputs adjustment actions. Through continuous interaction with the environment, the reinforcement learning model learns the optimal adjustment strategy under different abnormal conditions.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] In the advertising marketing system based on data update and organization, the data processing and analysis unit uses memory-based computing and parallel processing architecture, combined with adaptive window algorithms, asynchronous I / O technology and vectorized computing methods, to accurately calculate various indicators, deeply explore the potential value of data, accurately predict user behavior trends, and provide data support for advertising strategy formulation. The strategy generation module uses genetic algorithms to optimize the initial strategy and improve the scientific nature of the strategy. The strategy optimization module uses multi-agent deep reinforcement learning and experience playback technology to adjust parameters in real time. At the same time, the effect evaluation module considers the delay effect of advertising delivery to calculate the input-output ratio, so that the advertising strategy is more in line with market demand and the input-output ratio of advertising delivery is improved. The advertising delivery monitoring unit monitors the advertising conversion rate in real time, uses the autoencoder model to detect anomalies, and combines the reinforcement learning model to adjust the strategy in time to ensure that advertising delivery is always in the best state and improve the accuracy and efficiency of advertising marketing. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is an overall block diagram of the present invention.

[0042] The meaning of each number in the figure is:

[0043] 1. Data acquisition and transmission unit; 2. Data processing and analysis unit; 21. Data processing module; 22. Data analysis module; 3. Advertising strategy generation and optimization unit; 31. Strategy generation module; 32. Strategy optimization module; 33. Effect evaluation module; 4. Advertising delivery monitoring unit. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] The present invention provides an advertising and marketing system based on data update and organization, please refer to Figure 1 As shown, it includes a data collection and transmission unit 1, a data processing and analysis unit 2, an advertising strategy generation and optimization unit 3, and an advertising delivery monitoring unit 4;

[0046] The data collection and transmission unit 1 obtains user behavior data, market competition data and industry trend data through a data collector, and cleans and processes the collected data to obtain a comprehensive data set.

[0047] During data cleaning, the data collection and transmission unit 1 uses a density-based local outlier factor algorithm for outlier detection in user behavior data. The specific steps are as follows:

[0048] During the collection process, user behavior data will be interfered by various factors, resulting in outliers. These outliers may seriously affect the subsequent data processing and analysis results, leading to inaccurate advertising strategy generation. , calculate it to The distance between the nearest neighbors is is a pre-set parameter, denoted as ,in , Represents the nearest neighbor point, Expressed as nearest neighbor points, and take the The distance between neighboring points is No. distance .

[0049] User behavior data usually has complex distribution characteristics. The behavior patterns of different users may vary greatly. There may be multiple areas with different densities in the data set. The density-based local outlier factor algorithm does not rely on the specific distribution assumptions of the data and can accurately identify outliers in complex data distributions. By considering the local density information of data points, it can distinguish those points with large density differences from surrounding data points in the local area and determine them as outliers.

[0050] calculate of Distance from all points in the neighborhood to The average distance is denoted as The reachable distance is as follows: ;according to The reachable distance is calculated The local reachable density of ;in yes of Distance neighborhood, calculation The local outlier factor ;like If the local outlier factor of is greater than 1, then Determined as an outlier.

[0051] The LOF algorithm is a local density-based method that takes into account the local environment around each data point. The LOF algorithm can detect data points that behave abnormally in local areas, even if these points may not be obvious in the global distribution. It can accurately identify data points that behave abnormally in these niche groups without misjudging the entire niche group as abnormal, so that subsequent data processing and analysis results can more accurately reflect the user's true behavior patterns.

[0052] The data processing and analysis unit 2 uses a data processing module 21 based on memory computing and parallel processing architecture to process the data in the comprehensive data set, and uses an adaptive window algorithm to adjust the window size according to the real-time flow of the data to obtain the real-time change characteristics of the data. At the same time, it combines asynchronous I / O technology and vectorized computing methods to perform real-time aggregation analysis on the data, calculate user behavior indicators and market trend indicators, and store the user behavior indicators and market trend indicators in a graph database. The data analysis module 22 is then used to extract user behavior graphs and market relationship graphs from the graph database, and the user behavior graph and the market relationship graph are associated and mined through a graph neural network algorithm based on an attention mechanism to obtain a user behavior sequence, and then a long short-term memory network is used to model the user behavior sequence to predict future behavior trends.

[0053] In the data processing and analysis unit 2, when the data processing module 21 based on memory computing and parallel processing architecture adopts the adaptive window algorithm, the window size adjustment formula is as follows:

[0054] Before starting to apply the adaptive window algorithm, it is necessary to determine some necessary initial parameters. These parameters are the basis for subsequent calculation and adjustment of the window size. At the same time, start real-time monitoring of data traffic and record the current time Data flow In addition, define the adaptive coefficient , The value range is .

[0055] At each time step , get the current data real-time flow , and according to the flow at the last moment Calculate flow rate change The traffic change rate is a key indicator for measuring the real-time traffic changes of data. By calculating the traffic change rate, we can understand whether the data traffic is increasing, decreasing, or remaining stable, thus providing a basis for adjusting the window size.

[0056] The window size is adjusted according to the positive or negative situation of the calculated traffic change rate. When , it indicates that the data flow is increasing, and the window size is adjusted to ;when When the data flow is decreasing or remains unchanged, the window size is adjusted to The adaptive window size adjustment method can dynamically optimize the window size according to the actual changes in data traffic, which not only avoids the waste of computing resources caused by too large a window, but also avoids the problem of being unable to capture important data features due to too small a window.

[0057] In the data processing module 21, when the asynchronous I / O technology and the vectorized calculation method are combined to perform real-time aggregation analysis, the user behavior index is calculated as follows:

[0058] Records containing user ID, behavior time and behavior type are screened out from the comprehensive data set of the data acquisition and transmission unit 1 to form a user behavior data set. The user behavior data set may be very large. If the synchronous I / O technology is used to read the data sequentially, a lot of time will be spent waiting for the I / O operation to complete, resulting in low data processing efficiency. A programming language library that supports asynchronous I / O is used to read the user behavior data set in parallel. After reading the data, each record is converted into a vector form.

[0059] Asynchronous I / O technology and parallel reading greatly shorten the time of data reading and improve the overall efficiency of data processing. Vectorized data representation facilitates batch computing and can accelerate the subsequent data analysis process. The behavior type is discrete categorical data, and computers cannot directly process this text information. One-hot encoding is used to convert the behavior type into a vector form, which can convert categorical data into numerical data.

[0060] Define a collection of behavior types For each behavior type, create a vector with a length equal to the number of behavior types, in which only the position corresponding to the behavior type is 1 and the rest are 0. The behavior type is converted into a computable vector form, which provides a basis for the subsequent calculation of user behavior frequency indicators and behavior indicators.

[0061] The frequency of user behavior can reflect the user's activity level and behavior preferences within a certain period of time. By calculating the frequency indicators of different behavior types, we can understand the user's tendency towards different behaviors and provide an important reference for the formulation of advertising strategies. For each user, within a certain time window, The cumulative sum of different behavior type vectors is counted, and then the cumulative sum is divided by the time window length to obtain the behavior frequency vector. The behavior frequency index can quantify the user's behavior characteristics. By comparing the behavior frequency vectors of different users, the behavioral differences between users can be found, providing a basis for personalized advertising recommendations.

[0062] The behavior frequency vector is only a preliminary quantification of user behavior. By further calculating the user behavior index, we can comprehensively consider the importance of different behavior types, more comprehensively evaluate the user's behavior characteristics, and assign different weights to different behavior types. , let the behavior type set The corresponding weight vector is , then the user Behavioral indicators ;in is the behavior frequency vector No. elements.

[0063] In the data analysis module 22, the graph neural network algorithm based on the attention mechanism performs association mining on the user behavior graph and the market relationship graph in the following specific steps:

[0064] Before performing association mining, it is necessary to clarify the feature representation of each node in the user behavior graph and the market relationship graph. The node feature vector is the basis for subsequent calculations. Suppose the user behavior graph is ,in is a set of nodes in the user behavior graph. For each node , whose eigenvector is recorded as , the market relationship map is , for each node , whose eigenvector is recorded as ,These feature vectors ensure the accuracy and consistency of the subsequent ,association mining calculations, and avoid calculation errors or result deviations caused by unclear ,feature representations.

[0065] By linearly transforming the node feature vector to obtain the query vector, key vector and value vector, the node features can be mapped to different subspaces, so that the correlation between nodes can be measured from different angles when calculating the attention score. For each node in the user behavior graph , the query vector is obtained by linear transformation ,in is a learnable weight matrix, for each node in the market relationship graph , the key vector is obtained by linear transformation Sum value vector ,in and It is a learnable weight matrix. The introduction of a learnable weight matrix for linear transformation increases the flexibility and expressiveness of the model, and can automatically learn the importance of different node features in association mining.

[0066] The attention score is used to measure the correlation between the nodes in the user behavior graph and the nodes in the market relationship graph. and nodes in the market relationship graph , calculate the attention score, that is ;in is the dimension of the query vector and key vector, divided by This is to prevent the dot product result from being too large and to avoid the gradient vanishing problem when the softmax function is used later.

[0067] Input the attention score into the preset softmax function to calculate the attention weight ,The softmax function can normalize the attention scores, making the weights comparable, and highlighting the importance of nodes with higher correlation, which facilitates more reasonable weight allocation in subsequent calculations.

[0068] The associated feature vector integrates the information of each node in the market relationship graph. By weighted summing the value vector through the attention weight, the information of the market relationship graph nodes with high correlation with the user behavior graph nodes can be integrated into one vector, thereby mining the associated information between the two graphs.

[0069] For nodes in the user behavior graph , calculate the associated feature vector, the associated feature vector can effectively integrate the information related to the user behavior graph nodes in the market relationship graph, and concentrate the associated information scattered in multiple nodes into one vector, which is convenient for subsequent processing and analysis. After splicing the associated feature vector with the original node feature vector and undergoing nonlinear transformation, the mined associated information can be integrated into the characteristics of the user behavior graph nodes. The updated node feature vector contains more information about the relationship between the two graphs.

[0070] In the data analysis module 22, when the long short-term memory network models the user behavior sequence, the gated recurrent unit structure is optimized. The specific operations are as follows:

[0071] When processing user behavior sequences, traditional neural networks have difficulty capturing long-term dependencies in the sequence. The gated recurrent unit solves this problem by introducing a reset gate and an update gate. The reset gate controls how much information in the previous hidden state needs to be forgotten or reset, while the update gate determines how much new information is added to the current hidden state.

[0072] Assume that the input user behavior sequence is at time The vector is , the hidden state at the last moment , then the calculation formula for resetting the gate is , update gate ,in is the sigmoid function, The introduction of the weight matrix, reset gate and update gate enables the model to dynamically control the flow and update of information according to the input, avoiding the problem of gradient vanishing or gradient exploding in traditional recurrent neural networks and enhancing the model's ability to capture long-term dependencies.

[0073] Candidate hidden states The calculation formula is ,in, is the hyperbolic tangent function, which maps the input to the interval (-1,1); is a learnable weight matrix, Represents element-by-element multiplication. The use of the reset gate enables the model to flexibly decide which parts of the hidden state at the previous moment need to be retained or forgotten based on the input, thereby generating candidate hidden states that are more in line with the current input, improving the flexibility and adaptability of the model.

[0074] The final hidden state is calculated based on the update gate and the candidate hidden state. The update gate determines the proportion of the previous hidden state and the candidate hidden state in the final hidden state. The calculation formula is ;in, shows the retention ratio of the hidden state at the previous moment, It represents the integration ratio of candidate hidden states. The introduction of the update gate enables the model to dynamically balance the weights of historical information and current input information, avoiding the problem of over-reliance on historical information or current input information, and improving the stability and generalization ability of the model.

[0075] The final hidden state contains comprehensive information from the past to the current moment in the user behavior sequence. By analyzing and utilizing the final hidden state, we can mine the long-term dependencies in the user behavior sequence. The final hidden state of each moment is input as a feature into the subsequent analysis module. With the final hidden state as input and the user's actual behavior or other relevant indicators as output, we can build a model of the user behavior sequence to achieve more accurate user behavior prediction and analysis.

[0076] The advertising strategy generation and optimization unit 3 constructs a strategy generation engine through the strategy generation module 31, and generates an initial advertising delivery strategy in combination with real-time data analysis results and preset business rules. It also introduces a dynamic optimization algorithm based on multi-agent deep reinforcement learning through the strategy optimization module 32, and uses a combination of experience replay and priority experience replay to adjust strategy parameters in real time according to the effect of advertising delivery. At the same time, it establishes a real-time advertising effect evaluation index system, and uses the real-time calculation function of the effect evaluation module 33 to conduct real-time evaluation of the advertising input-output ratio.

[0077] In the advertisement strategy generation optimization unit 3, when the strategy generation module 31 constructs the strategy generation engine, a genetic algorithm is used to optimize the generation of the initial advertisement delivery strategy. The specific steps are as follows:

[0078] Assume that the advertising time can be divided into Time period, delivery channels In the different value levels, you can use a 1 to An integer representing a delivery channel, with a value from 1 to An integer representing the budget. The integer representation of , arrange these three integers in order to form an integer sequence of length 3 as a chromosome.

[0079] Random Generation Such chromosomes constitute the initial population ,in Indicates chromosomes, each chromosome represents an initial advertising strategy. The diverse initial population increases the search space of the algorithm and helps avoid falling into the local optimal solution.

[0080] For each chromosome in the initial population , according to the corresponding advertising delivery time, delivery channel and delivery budget, combined with historical data, calculate the expected conversion rate of advertising under this strategy , fitness function , i.e. chromosome The fitness value is equal to the expected conversion rate of the corresponding advertisement. Taking the expected conversion rate of advertisement as the fitness is directly related to the core goal of advertising marketing, and can guide the algorithm to evolve in the direction that is most conducive to improving the conversion rate, making the final advertising delivery strategy more valuable.

[0081] Calculate each chromosome The probability of selection ;in is the size of the population, and then generates a Random numbers in the interval , and accumulate the selection probability of each chromosome in turn ,when When The chromosomes are used as parent chromosomes, and this process is repeated to select a sufficient number of parent chromosomes for subsequent crossover and mutation operations. Through the selection operation, chromosomes with high fitness are retained and participate in reproduction, so that the overall fitness of the population is gradually improved, which is conducive to the algorithm to find a better advertising delivery strategy.

[0082] Randomly select two chromosomes from the selected parent chromosome and , randomly select an intersection point ,Will Before Genes and After 3- Gene combination, generate a sub-chromosome ,Will Before Genes and After 3- Gene combination, generating another daughter chromosome .

[0083] For the generated daughter chromosomes, The probability of mutation is used to mutate a gene in a chromosome, and a gene in the daughter chromosome is randomly selected and replaced with another random integer in the value range of the gene. The combination of crossover and mutation operations can not only make full use of the excellent genes of the parent chromosome, but also introduce new gene combinations, increasing the diversity and search space of the population, and helping the algorithm to jump out of the local optimal solution and find the global optimal solution.

[0084] The generated offspring chromosomes replace those chromosomes with the lowest fitness to form a new population. Repeat the above selection, crossover and mutation operation steps, calculate the fitness of each chromosome in the new population, select the parent chromosome, perform crossover and mutation operations, generate new offspring chromosomes, and repeat this process until the preset maximum number of iterations is reached. After multiple iterative optimizations, the chromosome with the highest fitness in the population is finally obtained, which is the optimized initial advertising delivery strategy.

[0085] The strategy optimization module 32 uses a method combining experience playback and priority experience playback to adjust strategy parameters in real time according to the advertising delivery effect. The specific operations are as follows:

[0086] During the advertising process, if there is Agents, at each time step , No. An agent will be in a state , based on the current strategy Select an action Perform advertising operations and set rewards based on the actual effect of advertising and the advertising conversion rate , and then the agent enters the next state , this set of information Stored as a sample in the experience replay pool Experience Replay Pool It is a first-in-first-out queue. When the queue reaches its maximum capacity, When the earliest sample is entered, it will be removed to facilitate a more comprehensive understanding of the advertising environment and thus optimize the advertising strategy more effectively.

[0087] Experience replay pool Each sample in Calculate a priority , the priority calculation is determined by the reward size, timestamp and prediction error. Let the reward weight be , the weight of the timestamp is , the weight of the prediction error is , the reward size is directly taken , the timestamp factor can be represented by a decay function, namely ,in is the current time, is the time when the sample enters the experience replay pool, is the decay coefficient, and the prediction error can be used to predict the reward of the sample using the deep reinforcement learning model. With actual rewards The absolute value of the difference is expressed as follows: ; Taking into account the reward size, timestamp and prediction error, it can more accurately evaluate the importance of each sample to model training, allowing the model to learn more targetedly.

[0088] Each time from the experience replay pool The sampling batch size is When a sample is selected, the priority sum of all samples in the experience replay pool is first calculated. , and then for each location that needs to be sampled, generate a Random numbers in the range , traverse the samples in the experience replay pool and calculate the cumulative priority ,when When samples as sampling results, repeat this process times, and get a batch of samples: ; It enables the model to learn key strategy information faster. By inputting the sampled samples into the deep reinforcement learning model for training, calculating the loss function and updating the model's parameters, the model can continuously learn and optimize the advertising delivery strategy.

[0089] In the real-time advertising effect evaluation index system established by the effect evaluation module 33, when calculating the advertising input-output ratio index, the delay effect of advertising delivery is taken into account. The specific formula is as follows:

[0090] Record the exact time when the ad starts running , and determine a time interval , used to calculate the revenue after advertising, and at the same time, to clarify the total cost incurred during the advertising process , which includes all expenses related to advertising, such as advertising production costs and delivery platform fees.

[0091] The time period after the ad is served Count each time point Benefits , according to time and ad delivery time The distance determines the weight Then, the delayed benefit is calculated as weighted average ; It makes the calculation of advertising input-output ratio more in line with the actual situation and provides a more accurate reference for the adjustment of advertising strategies.

[0092] Delayed return calculated based on and advertising costs , using the formula Calculate the advertising input-output ratio. The higher the ratio, the better the effectiveness of the advertising. Otherwise, the advertising strategy needs to be adjusted.

[0093] The advertising delivery monitoring unit 4 applies the optimized advertising strategy to the actual delivery, and monitors the conversion rate of the advertisement in real time. It uses the anomaly detection algorithm based on streaming time series analysis to monitor abnormal situations in the advertising delivery process, and analyzes the monitoring data in real time, and adjusts the advertising delivery strategy according to the analysis results.

[0094] In the advertising delivery monitoring unit 4, the anomaly detection algorithm based on streaming time series analysis adopts an autoencoder model, and the specific steps are as follows:

[0095] Assuming that the advertising conversion rate monitoring data is a time series, the minimum-maximum normalization method is used to normalize the data to interval, taking the normalized data as the input of the autoencoder, which improves the training efficiency and accuracy of the autoencoder and enables the model to better learn the characteristics of the advertising conversion rate data. The autoencoder consists of an encoder and a decoder.

[0096] The encoder maps the input to a low-dimensional representation, and the decoder reconstructs the low-dimensional representation into a new input for the autoencoder. The reconstruction error is defined by calculating the difference between the new input and the input of the autoencoder. The autoencoder is trained to minimize the reconstruction error. By minimizing the reconstruction error, the autoencoder can learn the main features and distribution of the data, achieve better reconstruction for normal data, and produce larger reconstruction errors for abnormal data, thereby providing an effective judgment basis for anomaly detection.

[0097] For new monitoring data, input it into the trained autoencoder to obtain the reconstructed input and calculate the reconstruction error. A threshold is set in advance. If the reconstruction error is greater than the threshold, the data point is determined to be an anomaly. Continue to monitor the advertising conversion rate data in real time. Once an anomaly is found, immediately mark and record the relevant information.

[0098] The anomaly detection results are input as states into the reinforcement learning model. The reinforcement learning model selects an adjustment action based on the current state. After the adjustment action is executed, the environment will feedback a reward. The reinforcement learning model continuously interacts with the environment, updates the strategy based on the reward signal, learns the optimal adjustment strategy under different abnormal states, and continuously optimizes the strategy to adapt to different abnormal scenarios, thereby improving the flexibility and effectiveness of advertising delivery.

[0099] In the present invention, multi-source data is collected and cleaned by the data collection and transmission unit 1 to provide a comprehensive data set. The data processing and analysis unit 2 uses memory computing, parallel processing architecture and multiple algorithms to calculate user behavior and market trend indicators, mine graph associations, and predict user behavior. The advertising strategy generation and optimization unit 3 builds a strategy generation engine to generate an initial strategy, uses deep reinforcement learning for dynamic optimization, and evaluates the advertising input-output ratio in real time. The advertising delivery monitoring unit 4 uses the optimized strategy for actual delivery, monitors the conversion rate in real time, detects anomalies through the autoencoder model and adjusts the strategy, thereby improving the accuracy and efficiency of advertising marketing, improving the input-output ratio of advertising delivery, and enhancing the advertising effect.

[0100] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. An advertising and marketing system based on data updating and sorting, characterized in that: It comprises a data collection and transmission unit (1), a data processing and analysis unit (2), an advertising strategy generation and optimization unit (3) and an advertising delivery monitoring unit (4); The data acquisition and transmission unit (1) acquires user behavior data, market competition data and industry trend data through a data collector, and cleans and processes the acquired data to obtain a comprehensive data set; The data processing and analysis unit (2) uses a data processing module (21) based on memory computing and parallel processing architecture to process the data in the comprehensive data set, uses an adaptive window algorithm to adjust the window size according to the real-time flow of the data, obtains the real-time change characteristics of the data, and at the same time, combines asynchronous I / O technology and vectorized computing methods to perform real-time aggregation analysis on the data, calculates user behavior indicators and market trend indicators, and stores the user behavior indicators and market trend indicators in a graph database, and then uses the data analysis module (22) to extract user behavior graphs and market relationship graphs from the graph database, and uses a graph neural network algorithm based on an attention mechanism to perform association mining on the user behavior graph and the market relationship graph to obtain a user behavior sequence, and then uses a long short-term memory network to model the user behavior sequence to predict future behavior trends; The advertising strategy generation optimization unit (3) constructs a strategy generation engine through a strategy generation module (31), and generates an initial advertising delivery strategy by combining real-time data analysis results and preset business rules, and introduces a dynamic optimization algorithm based on multi-agent deep reinforcement learning through a strategy optimization module (32), and uses a method combining experience replay and priority experience replay to adjust strategy parameters in real time according to the advertising delivery effect. At the same time, a real-time advertising effect evaluation index system is established, and the real-time calculation function of the effect evaluation module (33) is used to perform real-time evaluation of the advertising input-output ratio; The advertising delivery monitoring unit (4) applies the optimized advertising strategy to actual delivery, monitors the conversion rate of the advertisement in real time, monitors abnormal situations in the advertising delivery process using an anomaly detection algorithm based on streaming time series analysis, analyzes the monitoring data in real time, and adjusts the advertising delivery strategy according to the analysis results.

2. The advertising and marketing system based on data updating and arrangement according to claim 1, characterized in that: The data collection and transmission unit (1) adopts a density-based local outlier factor algorithm for outlier detection in user behavior data during data cleaning. The specific steps are as follows: For each data point in the user behavior dataset , calculate it to The distance between the nearest neighbors is is a pre-set parameter, denoted as ,in , Represents the nearest neighbor point, Expressed as nearest neighbor points, and take the The distance between neighboring points is No. distance ; calculate of Distance from all points in the neighborhood to The average distance is denoted as The reachable distance is based on The reachable distance is calculated The local reachable density of Locally reachable density calculation The local outlier factor of If the local outlier factor of is greater than 1, then Determined as an outlier.

3. The advertising and marketing system based on data updating and arrangement according to claim 2, characterized in that: In the data processing and analysis unit (2), when the data processing module (21) based on memory computing and parallel processing architecture adopts an adaptive window algorithm, the window size adjustment formula is as follows: Set the current window size to , the real-time data flow is The flow rate change rate is ,in Define an adaptive coefficient for the flow at the previous moment ; when When the window is resized to ;when When the window is resized to .

4. The advertising and marketing system based on data updating and arrangement according to claim 3 is characterized in that: In the data processing module (21), when the asynchronous I / O technology and the vectorized calculation method are combined to perform real-time aggregation analysis, the user behavior index is calculated as follows: Extracting a user behavior data set from the comprehensive data set of the data acquisition and transmission unit (1), wherein the user behavior data set includes a user ID, a behavior time, and a behavior type, reading the data in parallel using an asynchronous I / O technology, and converting the read data into a vector form; The behavior types include browsing, clicking and purchasing. For the behavior types, one-hot encoding is used for vectorization, and the user behavior frequency index is calculated. For each user, the cumulative sum of the different behavior type vectors within a certain time window is counted, and then divided by the length of the time window to obtain the behavior frequency vector, and the user behavior index is calculated based on the behavior frequency vector.

5. The advertising and marketing system based on data updating and arrangement according to claim 4 is characterized in that: In the data analysis module (22), when the graph neural network algorithm based on the attention mechanism performs association mining on the user behavior graph and the market relationship graph, the specific steps are as follows: Record and set the node feature vectors of the user behavior graph and the market relationship graph respectively. For each node in the user behavior graph, obtain the query vector through linear transformation. For each node in the market relationship graph, obtain the key vector and value vector. The attention score is calculated based on the query vector and the key vector, and the attention score is combined with the preset softmax function to calculate the attention weight. The associated feature vector is then calculated based on the attention weight and the value vector. The associated feature vector is concatenated with the original node feature vector and then subjected to a nonlinear transformation to obtain an updated node feature vector. The association relationship between the two graphs is mined based on the node feature vector.

6. The advertising and marketing system based on data updating and arrangement according to claim 5, characterized in that: In the data analysis module (22), when the long short-term memory network models the user behavior sequence, the gated recurrent unit structure is optimized, and the specific operation is as follows: For the input user behavior sequence and the previous hidden state , calculate the reset gate , update gate ,in is the sigmoid function, is a weight matrix, the reset gate is used to control the reset of information in the hidden state of the previous moment, and the update gate is used to control the transmission and update of information; The candidate hidden state is calculated based on the user behavior sequence, the hidden state at the previous moment, and the reset gate, and the final hidden state is calculated based on the candidate hidden state. Finally, the long-term dependency in the user behavior sequence is obtained and modeled based on the final hidden state.

7. The advertising and marketing system based on data updating and arrangement according to claim 6, characterized in that: In the advertising strategy generation optimization unit (3), when the strategy generation module (31) constructs the strategy generation engine, a genetic algorithm is used to optimize the generation of the initial advertising delivery strategy. The specific steps are as follows: Encode the advertising time, delivery channel and delivery budget into an integer sequence as a chromosome, and randomly generate a certain number of chromosomes as the initial population. Each chromosome represents an initial advertising delivery strategy, and then calculate the fitness based on the expected conversion rate of the advertisement. The parent chromosome is selected according to the fitness value, and the parent chromosome is crossover and mutation operations are performed to generate the child chromosome. The above steps are repeated until the maximum number of iterations is met to obtain the optimized initial advertising delivery strategy.

8. The advertising and marketing system based on data updating and arrangement according to claim 7, characterized in that: The strategy optimization module (32) uses a method combining experience playback and priority experience playback to adjust strategy parameters in real time according to the advertising delivery effect. The specific operations are as follows: During the advertising process, multiple agents interact with the advertising environment. Each agent goes through a state at each time step and selects an action based on the current strategy to perform advertising operations. It then receives a reward, which is set by the advertising conversion rate. It then enters the next state and stores this set of information as a sample in the experience replay pool, which is a first-in, first-out queue. When the queue is full, the earliest sample will be removed. A priority is calculated for each sample in the experience replay pool, where the priority is determined by the reward size, timestamp, and prediction error. Each time a batch of samples is sampled from the experience replay pool for training the deep reinforcement learning model, samples are sampled according to the priority of the samples to obtain a batch of samples, which are input into the deep reinforcement learning model for training. According to the deep reinforcement learning algorithm, the loss function is calculated and the parameters of the model are updated through the back propagation algorithm.

9. The advertising and marketing system based on data updating and arrangement according to claim 8, characterized in that: In the real-time advertising effect evaluation index system established by the effect evaluation module (33), when calculating the advertising input-output ratio index, the delay effect of advertising delivery is taken into account. The specific formula is as follows: Set the ad delivery time to , after the ad is placed Statistical income within time , the advertising cost is Considering that the revenue will be generated after a period of time after the advertisement is released, the delayed revenue is calculated by weighted average, that is: ； in yes The profit of the moment, It is based on time and ad delivery time The weight determined by the distance is .

10. The advertising and marketing system based on data updating and arrangement according to claim 9, characterized in that: In the advertising delivery monitoring unit (4), the anomaly detection algorithm based on streaming time series analysis adopts an autoencoder model, and the specific steps are as follows: Normalizing the monitoring data during the advertising delivery process and using it as the input of the autoencoder, wherein the monitoring data is the advertising conversion rate, and the autoencoder is composed of an encoder and a decoder; The encoder maps the input to a low-dimensional representation, and the decoder reconstructs the low-dimensional representation into a new input for the autoencoder. The reconstruction error is defined by calculating the difference between the new input and the input of the autoencoder, and the autoencoder is trained to minimize the reconstruction error. During actual detection, for new monitoring data, the reconstruction error is calculated. If the reconstruction error is greater than the preset threshold, the data point is determined to be an abnormal point and real-time monitoring is performed. Based on the abnormal detection results, the abnormal situation is input as a state into the reinforcement learning model, and the model outputs adjustment actions. Through continuous interaction with the environment, the reinforcement learning model learns the optimal adjustment strategy under different abnormal conditions.

Images