Budget allocation optimization method based on big data analysis

Abstract

The invention discloses a budget allocation optimization method based on big data analysis, and particularly relates to the technical field of budget management. According to the method, historical data of multi-channel budget expenditure is collected, a budget allocation basic data set is constructed, budget execution difference features are extracted, grouping identification is performed by using a clustering algorithm, and a key feature region is determined; extracting budget sensitivity features in combination with external economic index data; the budget sensitivity features are analyzed based on a random forest algorithm, and influence weights of the budget sensitivity features are determined; establishing a budget allocation optimization initial model, and optimizing a budget resource configuration parameter combination; generating a multi-scene simulation scheme of budget configuration, and evaluating a budget execution expected effect; and based on the budget execution expected effect and the preset budget execution efficiency threshold, selecting the optimal budget configuration scheme, and outputting the budget optimization allocation strategy, thereby realizing dynamic optimization of the budget allocation process, and being suitable for budget management in a complex multi-channel environment.

Description

Technical Field

[0001] The present invention relates to the technical field of budget management, and more specifically, to a method for optimizing budget allocation based on big data analysis. Background Art

[0002] In the existing budget allocation process, configuration generally relies on empirical rules, lacking in-depth analysis of multi-channel historical data and the ability to dynamically respond to changes in the external economic environment, making it difficult to identify key influencing factors in a timely manner and adjust budget resources. Traditional budget allocation methods often ignore the non-linear relationship between budget execution effects and external indicators, lacking the ability of intelligent mining and modeling based on big data, resulting in low efficiency and poor adaptability of budget allocation plans, and it is difficult to achieve the optimization of resource allocation. Summary of the Invention

[0003] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a method for optimizing budget allocation based on big data analysis to solve the problems raised in the above background art.

[0004] To achieve the above object, the present invention provides the following technical solutions: A method for optimizing budget allocation based on big data analysis, comprising the following steps: S1: Collect historical data of budget expenditures from multiple channels, construct a basic data set for budget allocation, and extract budget execution difference features; S2: Group and identify budget execution difference features based on a clustering algorithm to determine key feature regions; S3: Combine key feature regions and external economic indicator data to extract budget sensitivity features; S4: Analyze budget sensitivity features using a random forest algorithm to determine the influence weights of budget sensitivity features; S5: Establish an initial model for optimizing budget allocation based on the influence weights of budget sensitivity features, and optimize the parameter combination of budget resource allocation through a particle swarm algorithm; S6: Generate multi-scenario simulation plans for budget allocation based on the optimized parameter combination of budget resource allocation, and evaluate the expected budget execution effects of multiple scenarios; S7: Based on the expected budget execution effects of multiple scenarios and a preset budget execution efficiency threshold, select the optimal budget allocation plan and output a budget optimization allocation strategy.

[0005] In a preferred embodiment, S1 is specifically: Collect historical budget expenditure data from different budget expenditure channels; Arrange the historical budget expenditure data in chronological order to construct a basic data set for budget allocation; Calculate the difference data between the actual expenditure and the predetermined budget expenditure during the budget execution process based on the budget allocation basic data set; Calculate the budget execution difference characteristics based on the difference data.

[0006] In a preferred embodiment, S2 is specifically as follows: Perform cluster analysis based on the budget execution difference characteristics; Divide the budget execution difference characteristics into multiple cluster groups, and determine the cluster center and the budget execution difference characteristic data points within each cluster group; Calculate the Euclidean distance between the budget execution difference characteristic data points within each cluster group and the corresponding cluster center; Determine the dispersion degree of the budget execution difference characteristic data points according to the Euclidean distance; Define the cluster group with the largest dispersion degree as the key feature area.

[0007] In a preferred embodiment, S3 is specifically as follows: Obtain all the budget execution difference characteristic data points within the key feature area; Collect the external economic indicator data related to the budget execution; Perform data association processing on the budget execution difference characteristic data points and the external economic indicator data according to the corresponding time interval to generate a fusion data set; Perform correlation analysis on the fusion data set, calculate the correlation coefficient between the budget execution difference characteristic data points and the external economic indicator data, and determine the budget sensitivity characteristics.

[0008] In a preferred embodiment, S4 is specifically as follows: Based on the budget sensitivity characteristics, establish the input data set of the random forest algorithm; Perform random sampling on the input data set to generate several training data subsets, and use each training data subset to generate several decision trees respectively; Based on the generated several decision trees, perform feature importance analysis on the budget sensitivity characteristics respectively; Calculate the number of feature splits and the information gain value of the split nodes for the budget sensitivity characteristics in each decision tree; Determine the comprehensive importance score of the budget sensitivity characteristics according to the number of feature splits and the information gain value of the split nodes; Based on the comprehensive importance score of the budget sensitivity characteristics, determine the influence weight of the budget sensitivity characteristics.

[0009] In a preferred embodiment, S5 is specifically as follows: Construct an initial model for optimizing budget allocation according to the influence weight of the budget sensitivity characteristics; Optimize the initial model based on budget allocation to establish the search space of the particle swarm algorithm; Continuously perform iterative updates within the search space through the particle swarm algorithm; Set termination conditions for the particle swarm algorithm, and terminate the iteration when the maximum number of iterations is reached or the stable threshold condition of the overall historical optimal position of the particle swarm is satisfied; Output the combination of budget resource allocation parameters corresponding to the overall historical optimal position of the particle swarm as the optimized combination of budget resource allocation parameters.

[0010] In a preferred embodiment, S6 is specifically: Construct a set of multi-scenario simulation schemes for budget allocation based on the optimized combination of budget resource allocation parameters; Construct a simulation computing environment for the set of multi-scenario simulation schemes for budget allocation; Based on the simulation computing environment, calculate the expected budget execution effects corresponding to each budget expenditure channel under each multi-scenario simulation scheme; Evaluate each multi-scenario simulation scheme in the set of multi-scenario simulation schemes according to the expected budget execution effects, and output the evaluation indicators corresponding to each multi-scenario simulation scheme.

[0011] In a preferred embodiment, S7 is specifically: Compare the evaluation indicators of each multi-scenario simulation scheme with the preset budget execution efficiency threshold to determine whether each multi-scenario simulation scheme reaches the budget execution efficiency threshold; Select the multi-scenario simulation scheme with the highest evaluation indicator from the multi-scenario simulation schemes that reach the budget execution efficiency threshold and define it as the optimal budget allocation scheme; Output the combination of budget resource allocation parameters included in the optimal budget allocation scheme as the budget optimization allocation strategy.

[0012] The technical effects and advantages of a budget allocation optimization method based on big data analysis of the present invention: By collecting historical data on budget expenditures from multiple channels and constructing a basic data set for budget allocation, the accurate extraction of budget execution difference features is realized, and the comprehensiveness and basic accuracy of data analysis are improved; by using a clustering algorithm to identify key feature regions, the regions with large fluctuations in budget execution efficiency can be effectively located; by combining external economic indicators to extract budget sensitivity features, the response ability to external environmental changes is enhanced; by using a random forest algorithm to analyze the influence weights of budget sensitivity features, the accuracy of feature evaluation is improved; based on the influence weights of budget sensitivity features, an initial model for optimizing budget allocation is established and a particle swarm algorithm is used to optimize the budget resource allocation, thus enhancing the global optimal ability of the budget plan; through multi-scenario simulation and simulation evaluation, the budget plan is made to have dynamic adaptability; by comparing with the budget execution efficiency threshold to select the optimal plan, the decision-making optimization of budget allocation is realized, and the flexibility of budget allocation and the resource utilization efficiency are improved. Brief Description of the Drawings

[0013] Figure 1 It is a schematic diagram of an optimization method for budget allocation based on big data analysis according to the present invention. Detailed Embodiments

[0014] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described 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 of 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 shall fall within the protection scope of the present invention.

[0015] Embodiment Figure 1 An optimization method for budget allocation based on big data analysis according to the present invention is given, which includes the following steps: S1: Collect historical data on budget expenditures from multiple channels, construct a basic data set for budget allocation, and extract budget execution difference features; S2: Based on a clustering algorithm, group and identify budget execution difference features to determine key feature regions; S3: Combine key feature regions and external economic indicator data to extract budget sensitivity features; S4: Use a random forest algorithm to analyze budget sensitivity features to determine the influence weights of budget sensitivity features; S5: Establish an initial model for optimizing budget allocation according to the influence weights of budget sensitivity features, and optimize the parameter combination of budget resource allocation through a particle swarm algorithm; S6: Based on the optimized parameter combination of budget resource allocation, generate multi-scenario simulation plans for budget allocation, and evaluate the expected budget execution effects of multiple scenarios; S7: Based on the expected effects of budget execution in multiple scenarios and the preset budget execution efficiency threshold, select the optimal budget allocation plan and output the budget optimization and allocation strategy.

[0016] S1: Collect historical data on budget expenditures from multiple channels, construct a basic data set for budget allocation, and extract the characteristics of budget execution differences, including: Collect historical budget expenditure data from different budget expenditure channels; Different budget expenditure channels include, but are not limited to, search engine advertising channels, social media advertising channels, in - app advertising channels for mobile applications, and online video advertising channels, etc. The advertising platform collects historical budget expenditure data based on the historical budget expenditure data within different budget expenditure channels. The historical budget expenditure data includes the budget amount, the budget consumption time period, and the type of budget expenditure. The budget amount represents the total amount of funds actually invested in different advertising channels within a specific period. The budget consumption time period is the start date and end date when the budget actually occurs. For example, the budget is at the start date and end date of each month. The type of budget expenditure records the advertising form or advertising resource of the budget expenditure. For example, brand exposure - type advertisements and effect - oriented advertisements in the video advertising channel. The advertising platform obtains historical budget expenditure data from the background database or advertising management system of each advertising channel by automatically accessing the application programming interfaces (APIs) opened by each channel. The APIs opened by each channel are opened in the form of network interfaces and return data in a standard format; the advertising platform system parses and stores the data in a unified format according to its own unified data specifications to ensure the timeliness, accuracy, and effectiveness of data acquisition.

[0017] Arrange the historical budget expenditure data in chronological order to construct a basic data set for budget allocation; To ensure the continuity and completeness of the basic data set for budget allocation, the advertising platform sorts the historical budget expenditure data from different budget expenditure channels according to the start date and end date of budget expenditure. Specifically, the advertising platform arranges the budget expenditure data for each cycle of each channel in ascending order of the budget consumption time period to form a basic data set for budget allocation with a time - series structure. The basic data set for budget allocation generated through the sorting method ensures the temporal characteristics of the data and the integrity and accuracy of the data for each channel.

[0018] Calculate the difference data between the actual expenditure and the predetermined budget expenditure during the budget execution process based on the basic data set for budget allocation; The actual expenditure is the real expenditure that occurs in each time period for each channel as recorded by the advertising platform, and the pre-determined budget expenditure is the budget expenditure for the corresponding period planned in advance. The advertising platform calculates the actual expenditure based on the budget amount data recorded in the historical budget expenditure data. The pre-determined budget expenditure is obtained from the budget plan data set by the advertiser in advance. The above operations are repeated for each budget period of each channel to form a set of difference data. The set of difference data reflects the deviation degree between the actual budget execution situation and the advertiser's planned expectation.

[0019] Calculate the budget execution difference characteristics based on the difference data; The budget execution difference characteristics include the average value, variance value, and coefficient of variation value of the difference data. The advertising platform performs statistical calculations on the set of difference data for each corresponding period of each channel respectively: the calculation of the average value is the sum of the difference data of all periods within the channel divided by the number of periods; the calculation method of the variance value is the sum of the squares of the differences between the difference data of each period and the average value divided by the number of periods; the coefficient of variation value is the standard deviation of the set of difference data divided by the average value, where the standard deviation is the result after taking the square root of the variance value. The budget execution difference characteristics of each channel such as the search engine advertising channel and the social media advertising channel are calculated respectively in the above manner.

[0020] S2: Based on the clustering algorithm, group and identify the budget execution difference characteristics to determine the key feature areas, including: Conduct clustering analysis based on the budget execution difference characteristics; When conducting clustering analysis, the advertising platform selects the budget execution difference characteristics as the basic unit of clustering and classifies them by the similarity degree of the numerical characteristics of the budget execution difference characteristics. The advertising platform adopts the K-means clustering algorithm and through multiple iterative calculations, groups the budget execution difference characteristics with close numerical values into the same category, gradually realizing the aggregation and grouping of budget channels with similar budget execution effects.

[0021] Divide the budget execution difference characteristics into multiple clustering groups, and determine the clustering center and the data points of the budget execution difference characteristics within each clustering group; The clustering algorithm first makes a preliminary division of the budget execution difference features according to the numerical values of the budget execution difference features, so as to divide them into several different clustering groups. The budget execution difference feature data points within each clustering group have high similarity in numerical values. Taking the search engine advertising channel as an example, after division, one clustering group may contain budget execution difference feature data points with a lower average difference and a smaller variance; another clustering group may contain budget execution difference feature data points with a higher average difference and a larger variance. After division, the clustering centers are determined. The clustering center is the arithmetic mean of the numerical values of all budget execution difference feature data points within each clustering group, thus forming the clustering centers. Each budget execution difference feature data point belongs to the corresponding clustering group and is distributed around the corresponding clustering center.

[0022] Calculate the Euclidean distance between the budget execution difference feature data points within each clustering group and the corresponding clustering center; The calculation process of the Euclidean distance is as follows: for each clustering group, for each budget execution difference feature data point within the clustering group, calculate the difference between the budget execution difference feature data point and the clustering center of the clustering group, then square the difference and sum it up, and take the square root of the obtained sum to get the Euclidean distance between the budget execution difference feature data point and the corresponding clustering center. The above calculation process is repeated for each budget execution difference feature data point within each clustering group, so as to obtain a complete set of Euclidean distance data.

[0023] Determine the degree of dispersion of the budget execution difference feature data points according to the Euclidean distance; The degree of dispersion of the budget execution difference feature data points is measured by the mean of the Euclidean distances within the clustering group. The calculation method is to add up the Euclidean distance values of all budget execution difference feature data points within a clustering group and then divide by the total number of budget execution difference feature data points within the clustering group. The result obtained by calculation reflects the average degree of dispersion of the budget execution difference feature data points relative to the clustering center. The larger the value, the higher the degree of dispersion of the budget execution difference feature data points within the clustering group. By calculating the degree of dispersion of the budget execution difference feature data points of all clustering groups, a value reflecting the internal dispersion of all clustering groups is obtained, which is convenient for comparing and analyzing the differences in budget execution difference features between clustering groups.

[0024] Define the clustering group with the largest degree of dispersion as the key feature area; Through the degree of dispersion of the budget execution difference feature data points of all clustering groups calculated, determine the clustering group with the largest degree of dispersion, that is, the clustering group with the largest degree of dispersion value. Define the clustering group with the largest degree of dispersion as the key feature area. The key feature area includes all budget execution difference feature data points within the clustering group with the largest degree of dispersion and the corresponding clustering center.

[0025] S3: Combine the key feature regions and external economic indicator data to extract budget sensitivity features, including: Obtain all budget execution difference feature data points within the key feature regions; The key feature regions are the clustering groups with the highest degree of dispersion in the cluster analysis. The data points within the clustering groups include the budget execution difference features corresponding to different budget expenditure channels such as search engine advertising channels, social media advertising channels, in - mobile - application advertising channels, and online video advertising channels during corresponding periods.

[0026] Collect external economic indicator data related to budget execution; The external economic indicator data is economic indicator data collected by the advertising platform from external public data sources and may be related to the budget execution effect. The external economic indicator data includes, but is not limited to, multiple indicators reflecting the changing trend of the economic environment such as macro - economic growth rate data and inflation rate data. The collection period of the external economic indicator data is consistent with the corresponding period of the budget execution difference feature data to ensure the unity and correspondence in the time dimension of the data.

[0027] Perform data association processing on the budget execution difference feature data points and the external economic indicator data according to the corresponding time intervals to generate a fused data set; The data association processing method is to clarify the corresponding relationship between the budget execution difference feature data points and the external economic indicator data according to the time period. For example, the advertising platform associates the budget execution difference feature data points of a certain month for the search engine advertising channel with the corresponding macro - economic growth rate data of that month. Each data point in the fused data set contains the budget execution difference feature data and the external economic indicator data of the same period.

[0028] Perform a correlation analysis on the fused data set, calculate the correlation coefficient between the budget execution difference feature data points and the external economic indicator data, and determine the budget sensitivity features; Correlation analysis determines the correlation strength between the data points of budget execution difference features and external economic indicator data by fusing data sets. The calculation method of the correlation coefficient is as follows: Calculate the covariance between the values of the data points of budget execution difference features and the values of external economic indicator data respectively, then calculate the standard deviation of the data point values of budget execution difference features and the standard deviation of the data values of external economic indicator data, and divide the calculated covariance by the product of the two standard deviation values to obtain the correlation coefficient between the data points of budget execution difference features and external economic indicator data. For example, the advertising platform calculates the covariance between the average value of the budget execution difference features of the search engine advertising channel and the consumer confidence index data, calculates the standard deviations of the two types of data respectively, and then performs the above correlation coefficient calculation process. Repeat the above calculation process for all data points of budget execution difference features and external economic indicator data in the fused data set to obtain a complete set of correlation coefficients.

[0029] Set screening criteria to determine budget sensitivity features. The screening criteria are as follows: Select the data items with the standardized correlation coefficients in the top 20% of the entire set of standardized correlation coefficients, and define the corresponding data items as budget sensitivity features. For example, when the value of the standardized correlation coefficient calculated between the average value of the budget execution difference features of a certain advertising channel, such as the search engine advertising channel, and the industry prosperity index data is within the range of the top 20% of all standardized correlation coefficient sets, then the industry prosperity index data is one of the budget sensitivity features; if the standardized correlation coefficient corresponding to the consumer confidence index data is within the top 20% range, then the consumer confidence index data is also a budget sensitivity feature.

[0030] S4: Analyze the budget sensitivity features using the random forest algorithm to determine the influence weights of the budget sensitivity features, including: Establish an input data set for the random forest algorithm based on the budget sensitivity features; The advertising platform obtains the budget sensitivity features determined by screening. The budget sensitivity features include data items whose standardized correlation coefficients between budget expenditure channels such as search engine advertising channels, social media advertising channels, in - mobile - application advertising channels, and online video advertising channels and external economic indicator data exceed the screening criteria. The data items of the budget sensitivity features and the corresponding budget execution difference feature data constitute the input data set of the random forest algorithm. Completely collect and uniformly organize the budget sensitivity features of each budget channel in each budget cycle to form the input data set of the random forest algorithm.

[0031] Randomly sample the input data set to generate several training data subsets, and use each training data subset to generate several decision trees respectively; The random forest algorithm performs random sampling operations based on the input dataset. The random sampling uses a with-replacement method to draw data points from the input dataset. The number of data points drawn each time is the same as the number of the input dataset. By repeating the random sampling operation multiple times, multiple different training data subsets are obtained to ensure the difference and diversity between the datasets. Each training data subset is used to train and generate several decision trees. The generation method of the decision tree is based on the partitioning operation of the feature data in the training data subset. The decision tree algorithm is used to gradually partition the different feature dimensions of the data in the training data subset, thus forming a decision tree. Repeat the above training process so that each training data subset generates multiple decision trees, forming a random forest model.

[0032] Perform feature importance analysis on the budget sensitivity features respectively based on the generated several decision trees; The method of feature importance analysis is to calculate the number of times the budget sensitivity feature is used during node splitting in the decision tree and the contribution degree of the node splitting to the classification result. The contribution degree of the node splitting is measured by the information gain value.

[0033] Calculate the number of times of feature splitting and the information gain value of the splitting node for the budget sensitivity feature in each decision tree; Calculate the specific number of times the budget sensitivity feature is used for decision tree node splitting for each decision tree in the random forest model in turn. The number of node splitting times refers to the total number of times a certain budget sensitivity feature is explicitly used for data discrimination operations in a decision tree model. The calculation method of the information gain value is as follows: calculate the information entropy of the decision tree node set before splitting, then calculate the information entropy of the child nodes after node splitting, and calculate the difference in information entropy, which is the information gain value of the budget sensitivity feature during node splitting. Calculate the information gain value for all nodes in each decision tree to ensure the importance of each budget sensitivity feature is evaluated. For example, if the industry prosperity index data in the decision tree is used for node splitting multiple times and the information gain generated each time is relatively high, then the feature importance of the industry prosperity index data in the decision tree is relatively high.

[0034] Determine the comprehensive importance score of the budget sensitivity feature according to the number of times of feature splitting and the information gain value of the splitting node; The calculation method of the comprehensive importance score is as follows: normalize the number of times of node splitting of each budget sensitivity feature in all decision trees respectively, that is, divide the number of times of feature splitting by the total number of times of all feature splitting; normalize the information gain value of each budget sensitivity feature in all decision trees, that is, divide the information gain value of the feature by the total information gain value of all features; perform weighted summation on the normalized number of times of node splitting and the normalized information gain value respectively, and then add them to obtain the comprehensive importance score of the budget sensitivity feature.

[0035] Determine the influence weight of the budget sensitivity feature based on the comprehensive importance score of the budget sensitivity feature; Unify and normalize the comprehensive importance score of the budget sensitivity feature, and use the calculated value as the influence weight of the budget sensitivity feature.

[0036] S5: Establish an initial model for optimizing budget allocation based on the influence weight of the budget sensitivity feature, and optimize the parameter combination of budget resource allocation through the particle swarm algorithm, including: Construct an initial model for optimizing budget allocation based on the influence weight of the budget sensitivity feature; The initial model for optimizing budget allocation includes a parameter combination for budget resource allocation in two dimensions. The first dimension is the fund allocation ratio parameter for each budget expenditure channel, and the second dimension is the total budget allocation parameter. Each budget expenditure channel includes channels such as the search engine advertising channel, the social media advertising channel, the in - mobile - application advertising channel, and the online video advertising channel. The fund allocation ratio parameter is expressed as the ratio value of the corresponding fund investment in each budget channel to the total budget allocation. The total budget allocation parameter represents the total budget funds available for allocation to each channel by the advertiser during a specific budget execution period. When constructing the initial model for optimizing budget allocation, determine the initial value of the fund allocation ratio parameter for each channel based on the influence weight of the budget sensitivity feature. For example, when the influence weight of the industry prosperity index is relatively high, the fund allocation ratio parameter for the search engine channel is correspondingly increased. In this way, obtain the specific parameter combination of budget resource allocation as the initial model for optimizing budget allocation.

[0037] Based on the initial model for optimizing budget allocation, establish the search space of the particle swarm algorithm; The search space is defined as a multi - dimensional space composed of the parameter combination of budget resource allocation as dimensions. Each dimension represents the possible values of the fund allocation ratio parameter for the budget expenditure channel and the total budget allocation parameter. Each particle in the search space represents a parameter combination of budget resource allocation, that is, a potential budget allocation solution. For example, a potential solution corresponding to the position of a certain particle is that the fund allocation ratio for the search engine channel is 20%, the fund allocation ratio for the social media channel is 30%, the fund allocation ratio for the in - mobile - application channel is 30%, the fund allocation ratio for the online video channel is 20%, and the total budget allocation parameter is a specific value. The advertising platform records and tracks the specific position of each particle in the search space to distinguish different potential budget allocation solutions.

[0038] Continuously perform iterative updates within the search space through the particle swarm algorithm; Determine the parameter settings for the particle swarm algorithm. First, set the number of particles, that is, the total number of particles in the search space of the particle swarm algorithm. The number of particles is determined through multiple experiments before the initial operation of the algorithm to ensure the effectiveness and stability of the algorithm operation. The initial position is the initial numerical value of the budget resource allocation parameter combination corresponding to each particle in the particle swarm algorithm. The initial position is based on the initial model of budget allocation optimization and is adjusted to make the initial positions of each particle evenly distributed. The initial velocity is the change rate of the initial movement state of each particle in the search space, and the initial velocity value is set by randomly taking values within a small range.

[0039] The inertia weight is used to control the degree to which particles in the particle swarm algorithm are affected by their own inertia during movement. The individual learning factor represents the degree to which a particle is affected by its own historical best position, and the global learning factor represents the degree to which a particle is affected by the historical best position of the entire particle swarm. To determine the values of the inertia weight, individual learning factor, and global learning factor, multiple groups of experiments and data tests are pre-conducted, and the values that make the algorithm have the best convergence effect and the fastest convergence speed are selected. The value range of the inertia weight is between zero and one, and the value ranges of the individual learning factor and the global learning factor are between one and three to ensure the balance of the particle swarm algorithm during the search process.

[0040] Specifically, the iterative update process of the particle swarm algorithm includes continuous updates in two dimensions: position and velocity. In each iteration of the algorithm, the updated velocity and new position of the particle are calculated separately. The velocity update calculation method is the product of the inertia weight value of the particle and the current velocity of the particle, plus the product of the individual learning factor and the difference between the particle's own historical optimal position, plus the product of the global learning factor and the difference between the historical optimal position of the entire particle swarm, to form the new velocity of the particle. The position update calculation method is the sum of the original position value of the particle and the new velocity value of the particle to determine the new position of the particle. Record the historical optimal position of each particle and track and record the historical optimal position of the entire particle swarm in real time for the next iterative calculation.

[0041] Set termination conditions for the particle swarm algorithm. When the maximum number of iterations is reached or the stable threshold condition of the historical optimal position of the entire particle swarm is satisfied, terminate the iteration; Set the conditions for terminating the iteration of the particle swarm algorithm, including the maximum number of iterations condition and the stable threshold condition of the historical optimal position of the whole. The maximum number of iterations is the total number of iterative calculations of the algorithm. The stable threshold condition of the historical optimal position of the whole particle swarm is that when the change amplitude of the historical optimal position of the whole particle swarm in consecutive multiple iterations is less than the preset change amplitude threshold, the algorithm is terminated. After each iteration, judge whether the algorithm satisfies one of the termination conditions. When the termination condition is satisfied, immediately stop the algorithm from continuing to iterate.

[0042] Output the parameter combination of the budget resource allocation corresponding to the overall historical optimal position of the particle swarm as the optimized parameter combination of the budget resource allocation; When the particle swarm algorithm iteration terminates, output the parameter combination of the budget resource allocation corresponding to the overall historical optimal position of the particle swarm as the optimized parameter combination of the budget resource allocation.

[0043] S6: Based on the optimized parameter combination of the budget resource allocation, generate multi-scenario simulation plans for the budget allocation, and evaluate the expected budget execution effects of multiple scenarios, including: Construct a set of multi-scenario simulation plans for the budget allocation based on the optimized parameter combination of the budget resource allocation; The parameter combination of the budget resource allocation includes the fund allocation ratio parameters and the total budget allocation parameters for each budget expenditure channel. Each budget expenditure channel includes the search engine advertising channel, the social media advertising channel, the in-mobile application advertising channel, and the online video advertising channel. Based on the parameter combination of the budget resource allocation, the advertising platform sets multiple scenarios, which are generated by adjusting different combinations of the fund allocation ratio parameters and the total budget allocation parameters. For example, in the first simulation plan scenario, a higher proportion of funds is allocated to the search engine advertising channel and a lower proportion of funds is allocated to the social media advertising channel; in the second simulation plan scenario, the fund allocation ratio of the online video advertising channel is increased and the fund ratios of other channels are correspondingly decreased; in the third simulation plan scenario, the fund allocation ratios of each channel are evenly distributed. In the above way, the advertising platform constructs multiple scenarios to form a set of multi-scenario simulation plans for the budget allocation for simulation calculation and analysis.

[0044] Construct a simulation calculation environment for the set of multi-scenario simulation plans for the budget allocation; The advertising platform constructs a simulation calculation environment for the set of multi-scenario simulation plans to evaluate the expected budget execution effects of each scenario plan. The simulation calculation environment includes the time dimension of the budget expenditure, the budget expenditure channel dimension, and the budget sensitivity characteristic dimension. The time dimension of the budget expenditure represents the periodic characteristics of the budget input and expenditure, such as one month or one quarter; the budget expenditure channel dimension includes the search engine advertising channel, the social media advertising channel, the in-mobile application advertising channel, and the online video advertising channel; the budget sensitivity characteristic dimension represents the external economic indicators that affect the budget execution effect. The advertising platform determines the parameters of the simulation calculation environment, such as the fund usage efficiency indicators of the budget expenditure channels and the expected influence degree of the budget sensitivity characteristics on the budget execution effect, and forms the input data of the simulation environment.

[0045] Calculate the expected budget execution effects corresponding to each budget expenditure channel under each multi-scenario simulation plan based on the simulation calculation environment; In the simulation computing environment, perform simulation calculations on each scenario in the multi-scenario simulation plan set for budget allocation to obtain the expected budget execution effects of each budget expenditure channel. The expected budget execution effect is expressed as the expected deviation value between the actual budget expenditure situation and the expected budget execution target. The calculation process is as follows: Based on the fund allocation ratio parameter and the total budget allocation parameter of the budget expenditure channel, calculate the expected value of the actual expenditure funds for each channel; Based on the influence weight of the budget sensitivity characteristics, calculate the influence of external economic indicators on the budget execution effect. Specifically, multiply the fund allocation ratio parameter by the influence weight of the economic indicator, and then combine it with the total budget allocation parameter to determine the expected expenditure effect of the budget execution for each channel. For example, in a certain scenario plan, when the fund allocation ratio of the search engine advertising channel is relatively high and the influence weight of the industry prosperity index data is relatively large, the expected budget execution effect corresponding to the search engine advertising channel is relatively good. Repeat the above calculation process for each scenario plan to obtain a complete set of expected budget execution effect data.

[0046] Evaluate each multi-scenario simulation plan in the multi-scenario simulation plan set according to the expected budget execution effect, and output the evaluation indicators corresponding to each multi-scenario simulation plan; Evaluate each simulation plan in the multi-scenario simulation plan set according to the set of expected budget execution effect data. The evaluation indicators include the budget usage efficiency indicator, the fund usage accuracy indicator, and the comprehensive utilization effect indicator of budget resources. The budget usage efficiency indicator is expressed as the expected value of the advertising effect brought by the budget expenditure funds; The fund usage accuracy indicator represents the gap between the expected value of the actual budget expenditure and the expected target budget expenditure value; The comprehensive utilization effect indicator of budget resources represents the degree of adaptation between the overall allocation of budget resources and external economic indicators. The calculation method of the evaluation indicators is as follows: The budget usage efficiency indicator is the average value of the expected budget execution effect values of each budget channel; The fund usage accuracy indicator is the absolute value of the difference between the expected value of the actual budget expenditure and the target budget expenditure value; The comprehensive utilization effect indicator of budget resources is the weighted average of the degree of adaptation between the expected budget execution effect and external economic indicators, and the weight is determined by the influence weight of the budget sensitivity characteristics. For example, when the value of the budget usage efficiency indicator is relatively high, the value of the fund usage accuracy indicator is relatively small, and the value of the comprehensive utilization effect indicator of budget resources is relatively good in a certain simulation plan scenario, mark this plan as an optimal plan. The advertising platform evaluates all multi-scenario simulation plans respectively, and records and outputs the evaluation indicators corresponding to each multi-scenario simulation plan.

[0047] S7: Based on the expected budget execution effect of multiple scenarios and the preset budget execution efficiency threshold, select the optimal budget allocation plan and output the budget optimization allocation strategy, including: Compare the evaluation indicators of each multi-scenario simulation plan with the preset budget execution efficiency threshold to determine whether each multi-scenario simulation plan meets the budget execution efficiency threshold; The set of evaluation indicators includes the budget utilization efficiency indicator, the capital utilization accuracy indicator, and the comprehensive utilization effect indicator of budget resources. The budget execution efficiency threshold is preset in advance. The budget execution efficiency threshold is expressed as the minimum acceptance standard for the ratio of the advertising effect brought by the budget expenditure to the invested funds, and the determination method of the budget execution efficiency threshold is a numerical standard set by referring to historical data and business objectives. For example, the budget execution efficiency threshold can be expressed as that the budget utilization efficiency indicator must reach a specific numerical standard, the gap value of the capital utilization accuracy indicator shall not exceed a certain limit, and the comprehensive utilization effect indicator of budget resources should be higher than a specific standard value. Compare the numerical values of the budget utilization efficiency indicator, the capital utilization accuracy indicator, and the comprehensive utilization effect indicator of budget resources with the specific standards in the corresponding budget execution efficiency threshold one by one, record the differences between each indicator of each simulation plan and the budget execution efficiency threshold, and determine whether the plan meets the requirements of the budget execution efficiency threshold. Through the above comparison process, all budget allocation multi-scenario simulation plans that reach the budget execution efficiency threshold are screened out.

[0048] Select the multi-scenario simulation plan with the highest evaluation indicator from the multi-scenario simulation plans that reach the budget execution efficiency threshold and define it as the optimal budget allocation plan; Sort each simulation plan that meets the budget execution efficiency threshold. The sorting basis is the comprehensive evaluation indicator of the budget utilization efficiency indicator, the capital utilization accuracy indicator, and the comprehensive utilization effect indicator of budget resources. The calculation method of the comprehensive evaluation indicator is that after the numerical values of the budget utilization efficiency indicator, the capital utilization accuracy indicator, and the comprehensive utilization effect indicator of budget resources are respectively standardized, weight distribution is carried out and then added together. Among them, the weight of the budget utilization efficiency indicator is the largest, the weight of the capital utilization accuracy indicator is the second, and the weight of the comprehensive utilization effect indicator of budget resources is the smallest. The weight values are determined through the analysis of historical data within the advertising platform and market experience. Through sorting, the multi-scenario simulation plan corresponding to the highest comprehensive evaluation indicator among all simulation plans that reach the budget execution efficiency threshold is obtained. The multi-scenario simulation plan with the highest comprehensive evaluation indicator is defined as the optimal budget allocation plan.

[0049] Output the combination of budget resource allocation parameters included in the optimal budget allocation plan as the budget optimization allocation strategy; The optimal budget allocation plan includes specific combinations of budget resource allocation parameters, including the fund allocation ratio parameters for each budget expenditure channel and the total budget allocation parameter. Each budget expenditure channel includes search engine advertising channels, social media advertising channels, in-mobile application advertising channels, and online video advertising channels, etc. The combination of budget resource allocation parameters represents the fund ratio and total fund allocation for each channel in the optimal budget allocation plan.

[0050] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters and threshold selection in the formulas are set by those skilled in the art according to the actual situation.

[0051] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wire (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more collections of available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0052] Those of ordinary skill in the art can realize that the modules and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0053] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and modules described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0054] In several embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods. For example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the devices or modules can be in electrical, mechanical, or other forms.

[0055] The modules described as separate components may or may not be physically separated. The components shown as modules may or may not be physical modules. They can be located in one place or distributed to multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0056] In addition, in each embodiment of the present application, the functional modules can be integrated in a processing module, or each module can exist physically alone, or two or more modules can be integrated in one module.

[0057] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art or part of this technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application. The foregoing storage medium includes various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0058] As described above, it is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims described above.

[0059] Finally: The above description is only the preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An optimization method for budget allocation based on big data analysis, characterized in that, The steps include: S1: Collect historical data on budget expenditures from multiple channels, build a basic data set for budget allocation, and extract budget execution difference characteristics; S2: Group and identify budget execution difference features based on clustering algorithms to determine key feature areas; S3: Combine key feature areas and external economic indicator data to extract budget sensitivity features; S4: Use the random forest algorithm to analyze the budget sensitivity characteristics and determine the impact weight of the budget sensitivity characteristics; S5: Establish an initial budget allocation optimization model based on the influence weights of budget sensitivity characteristics, and optimize the parameter combination of budget resource allocation through particle swarm optimization; S6: Based on the optimized parameter combination of budget resource allocation, generate multi-scenario simulation plans for budget allocation and evaluate the expected budget execution effects of multiple scenarios; S7: Based on the expected budget execution effects in multiple scenarios and the preset budget execution efficiency threshold, the optimal budget configuration plan is selected and the budget optimization allocation strategy is output.

2. The budget allocation optimization method based on big data analysis according to claim 1, wherein S1, specifically: Collect historical budget expenditure data from different budget expenditure channels; Arrange historical budget expenditure data in chronological order to construct a basic dataset for budget allocation; Calculate the difference between actual expenditure and planned budget expenditure during budget execution based on the budget allocation basic data set; Calculate budget execution variance characteristics based on variance data.

3. An optimization method for budget allocation based on big data analysis according to claim 2, characterized in that S2, specifically: Conduct cluster analysis based on budget execution difference characteristics; Dividing the budget execution difference characteristics into multiple cluster groups, and determining a cluster center and budget execution difference characteristic data points within each cluster group; Calculate the Euclidean distance between the budget execution difference feature data point and the corresponding cluster center within each cluster group; Determine the dispersion of budget execution difference feature data points based on Euclidean distance; The cluster group with the largest dispersion is defined as the key feature area.

4. The budget allocation optimization method based on big data analysis according to claim 3, wherein S3, specifically: Obtain all budget execution difference feature data points within the key feature area; Collect data on external economic indicators related to budget execution; Perform data association processing on the budget execution difference feature data points and external economic indicator data according to the corresponding time interval to generate a fused data set; Perform correlation analysis on the fused data set, calculate the correlation coefficient between the budget execution difference characteristic data points and the external economic indicator data, and determine the budget sensitivity characteristics.

5. The optimization method for budget allocation based on big data analysis according to claim 4, characterized in that, S4, specifically: Establish the input dataset of the random forest algorithm based on the budget sensitivity feature; Randomly sample the input data set to generate several training data subsets, and use each training data subset to generate several decision trees; Based on the generated decision trees, feature importance analysis is performed on the budget sensitivity features respectively; Calculate the number of feature splits and the information gain value of the split nodes for the budget sensitivity feature in each decision tree; The comprehensive importance score of the budget sensitivity feature is determined based on the number of feature splits and the information gain value of the split node; Based on the comprehensive importance score of budget sensitivity features, the impact weight of budget sensitivity features is determined.

6. The optimization method for budget allocation based on big data analysis according to claim 5, characterized in that S5, specifically: Construct an initial model for budget allocation optimization based on the impact weights of budget sensitivity characteristics; Optimize the initial model based on budget allocation to establish the search space of the particle swarm algorithm; Continuously perform iterative updates within the search space through the particle swarm algorithm; Set termination conditions for the particle swarm algorithm. When the maximum number of iterations is reached or the stable threshold condition of the overall historical optimal position of the particle swarm is satisfied, terminate the iteration; Output the combination of budget resource allocation parameters corresponding to the overall historical optimal position of the particle swarm as the optimized budget resource allocation parameter combination.

7. An optimization method for budget allocation based on big data analysis according to claim 6, characterized in that, S6, specifically: Construct a set of multi-scenario simulation schemes for budget allocation based on the optimized budget resource allocation parameter combination; Build a simulation computing environment for the set of multi-scenario simulation schemes for budget allocation; Based on the simulation computing environment, calculate the expected budget execution effects corresponding to each budget expenditure channel under each multi-scenario simulation scheme; Evaluate each multi-scenario simulation scheme in the set of multi-scenario simulation schemes according to the expected budget execution effects, and output the evaluation indicators corresponding to each multi-scenario simulation scheme.

8. An optimization method for budget allocation based on big data analysis according to claim 7, characterized in that S7, specifically: Compare the evaluation indicators of each multi-scenario simulation scheme with the preset budget execution efficiency threshold to determine whether each multi-scenario simulation scheme reaches the budget execution efficiency threshold; Select the multi-scenario simulation scheme with the highest evaluation indicator from the multi-scenario simulation schemes that reach the budget execution efficiency threshold, and define it as the optimal budget allocation scheme; Output the combination of budget resource allocation parameters included in the optimal budget allocation scheme as the budget optimization allocation strategy.

Images