Multi-platform internet marketing activity effect evaluation and prediction model construction method

Abstract

The invention provides a multi-platform internet marketing activity effect evaluation and prediction model construction method, and relates to the technical field of data analysis. The method comprises the following steps: collecting marketing activity data from a plurality of internet platforms; performing cleaning, normalization and feature engineering processing on the data; constructing a multi-level evaluation and prediction model based on a machine learning algorithm, wherein the multi-level evaluation and prediction model comprises a data fusion model, an effect evaluation model and an effect prediction model; calculating a click rate, a conversion rate, a user retention rate and a user life cycle value by using an effect evaluation model so as to comprehensively evaluate a marketing activity effect; and performing effect prediction by using the effect prediction model, and evaluating the model through a root-mean-square error, a mean absolute error and a mean absolute percentage error. According to the invention, the accuracy and comprehensiveness of marketing activity effect evaluation and prediction can be improved, and an important basis is provided for optimizing a marketing strategy.

Description

Technical Field

[0001] The present invention relates to the field of data analysis technology, and in particular to a method for constructing a multi-platform Internet marketing activity effect evaluation and prediction model. Background Art

[0002] With the rapid development and popularization of Internet technology, Internet marketing has become an important means for enterprises to promote their products and services. Multi-platform Internet marketing covers multiple channels such as search engines, social media, email marketing, content marketing, etc., which can widely reach potential customers and achieve precision marketing. However, in the face of massive multi-platform data, how to effectively evaluate and predict the effects of marketing activities has become a major challenge facing enterprises.

[0003] Traditional marketing effectiveness evaluation methods often rely on data analysis of a single platform, such as focusing only on the performance of a certain search engine or social media platform. This method not only fails to fully reflect the overall marketing effect, but also easily ignores the data interaction and synergy between different platforms. In addition, traditional methods often use simple statistical analysis methods, lack the ability to deeply mine and predict data, and are difficult to cope with the complex and changing market environment.

[0004] Existing technologies have many deficiencies in data preprocessing and feature engineering. Data preprocessing is often just a simple missing value filling and outlier removal, which fails to fully consider the diversity and complexity of data. In feature engineering, existing methods are mostly limited to manually selecting features, which makes it difficult to efficiently extract potential information from data. Normalization methods usually use Z-score standardization, which cannot handle uneven data distribution.

[0005] In terms of effect evaluation and prediction model construction, traditional methods mostly rely on simple models such as linear regression and decision trees, which cannot capture nonlinear relationships in data. With the development of machine learning and deep learning technologies, models such as long short-term memory networks (LSTM) and random forests have gradually been applied to marketing data analysis, improving prediction accuracy. However, existing technologies still have much room for improvement in model fusion and multi-step prediction strategies. Summary of the invention

[0006] In order to solve the technical problems in the prior art, the present invention provides a method for constructing a multi-platform Internet marketing activity effect evaluation and prediction model.

[0007] The technical solution provided by the present invention is as follows:

[0008] The present invention provides a method for constructing a multi-platform Internet marketing activity effect evaluation and prediction model, comprising:

[0009] S1. Collect marketing activity related data from multiple Internet platforms, including exposure, click-through rate, conversion rate, interaction volume and sales data;

[0010] S2. Clean, normalize and feature engineer the collected data to generate a unified data format for subsequent analysis;

[0011] S3. Based on machine learning algorithms, build multi-level evaluation and prediction models, including data fusion models, effect evaluation models, and effect prediction models;

[0012] S4. Evaluate the effect of the marketing campaign using the effect evaluation model, and use the following formula to calculate the click-through rate (CTR), conversion rate (CVR), user retention rate (RR) and customer lifetime value (CLV) to comprehensively evaluate the effect of the marketing campaign;

[0013] S5. Use the effect prediction model constructed based on historical data to perform effect prediction.

[0014] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0015] (1) In the present invention, the marketing campaign effect is comprehensively evaluated through multi-dimensional indicator calculation and comprehensive score calculation. Key performance indicators such as click-through rate, conversion rate, user retention rate and user lifetime value are used, combined with fuzzy comprehensive evaluation method, so that the evaluation results are more accurate and comprehensive, which is helpful for the optimization and decision-making of marketing strategies;

[0016] (2) In the present invention, a long short-term memory network (LSTM) model is used to improve the performance of the effect prediction model through feature engineering processing, model selection and training, multi-step prediction strategy and other steps. Through deep learning and analysis of historical data, future marketing effects can be accurately predicted, helping enterprises to carry out effective market planning and resource allocation;

[0017] (3) In the present invention, during the data processing process, the data quality and consistency are ensured by removing missing values ​​and outliers, adopting the Min-Max normalization method, standardization processing and weighted fusion, and using the exponential smoothing method to reduce data noise, making the data fusion model more robust and reliable, providing high-quality data support for subsequent evaluation and prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 A flowchart of a method for building a multi-platform Internet marketing campaign effect evaluation and prediction model is provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0021] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0022] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same. "of", "corresponding, relevant" and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same.

[0023] In the embodiments of the present invention, sometimes a subscript such as W1 may be mistakenly written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are consistent.

[0024] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0025] Reference Manual Attached Figure 1 , showing a flow chart of a method for building a multi-platform Internet marketing activity effect evaluation and prediction model provided by an embodiment of the present invention.

[0026] The embodiment of the present invention provides a method for building a multi-platform Internet marketing activity effect evaluation and prediction model, and the processing flow may include the following steps:

[0027] S1. Collect marketing activity related data from multiple Internet platforms, including exposure, click-through rate, conversion rate, interaction volume and sales data.

[0028] It is understandable that the Internet platform for collecting data needs to be determined first, including but not limited to the following: search engine advertising platform, social media platform, email marketing platform, content marketing platform and e-commerce platform. In order to comprehensively evaluate the effectiveness of marketing activities, the following key indicators need to be collected: Impressions: the number of times an advertisement or content is displayed in front of users; Click-Through Rate (CTR): the ratio of the number of times a user clicks on an advertisement or content to the number of impressions; Conversion Rate (CVR): the ratio of the number of times a user completes a predetermined conversion behavior (such as purchase, registration, etc.) to the number of clicks; Engagements: the number of times a user interacts with an advertisement or content, including likes, comments, shares, etc.; Sales Data: the number of sales and orders generated directly through marketing activities. According to the characteristics of different platforms and the data interfaces provided, it is necessary to select the appropriate data collection method. For platforms that provide API interfaces, you can write programs to automatically obtain the required data. For platforms that do not provide API interfaces, use Web crawling technology to obtain data. Some platforms also allow users to manually export data. Common formats such as CSV and Excel can be exported and imported into the system regularly. In order to facilitate subsequent analysis and processing, the collected data needs to be reasonably stored and managed, including data format unification, data storage and data backup. The frequency of data collection should be determined according to business needs.

[0029] S2. Clean, normalize and feature engineer the collected data to generate a unified data format for subsequent analysis.

[0030] It should be noted that the main purpose of data cleaning is to improve data quality and ensure the accuracy and reliability of analysis results.

[0031] Missing values ​​and outliers were removed, and the data were transformed to the [0,1] interval using the Min-Max normalization method.

[0032] In one possible implementation, missing values ​​are removed. Missing values ​​refer to the lack of data in certain fields of some records in the data set. The processing of missing values ​​is crucial to ensure data integrity and analysis accuracy. For numerical data, the following method is used to fill missing values:

[0033]

[0034] In the above formula, the upper and lower terms are respectively iWhen the value is non-missing and when x i When x is a missing value, i represents the i-th data point, and n represents the number of non-missing data points. The purpose of this processing is to make each field of the data set have complete information to avoid analysis errors caused by missing values.

[0035] In one possible implementation, outliers are removed. Outliers refer to data points that are significantly different from most of the data, usually due to measurement errors, data entry errors, or other unexpected reasons. The outliers are identified using a box plot method, and the range of outliers is determined by calculating the interquartile range (IQR):

[0036] IQR=Q3-Q1

[0037] LSL=Q1-1.5×IQR

[0038] USL=Q3+1.5×IQR

[0039] In the above formula, Q1 and Q3 represent the first quartile and the third quartile respectively, LSL represents the lower limit, and USL represents the upper limit. Data exceeding the upper and lower limits are regarded as outliers and are processed by the following method:

[0040]

[0041] In this way, the interference of extreme values ​​on data analysis can be effectively removed to ensure the reliability of data.

[0042] In a possible implementation, the main purpose of data normalization is to convert data of different dimensions to the same scale, so that each feature has equal importance in analysis and modeling, and to avoid deviations caused by dimensional differences. The processed data is normalized by Min-Max and converted to the [0,1] interval. The formula is as follows:

[0043]

[0044] In the above formula, x i represents the original data, x′ i represents the normalized data, x min and x max They represent the minimum and maximum values ​​of the data respectively. In this way, the influence of dimension on data analysis can be eliminated, so that the data of all features are distributed in the same interval, which is convenient for subsequent machine learning algorithm processing.

[0045] S3. Based on machine learning algorithms, build multi-level evaluation and prediction models, including data fusion models, effect evaluation models and effect prediction models.

[0046] It should be noted that the construction of the data fusion model aims to integrate data from multiple Internet platforms into a comprehensive indicator to provide a comprehensive marketing campaign effect. The specific construction method of the data fusion model includes: standardizing the data of each platform to make it in the same dimension, using the following formula:

[0047]

[0048] In the above formula, D′ ij represents the standardized data, D ij represents the original data, μ j and σ j Represent the mean and standard deviation of the jth platform respectively. Data standardization converts the data of different platforms into the same dimension, making the data of different platforms comparable. Since the data scale and distribution of each platform may be quite different, standardization can eliminate these differences, making the data suitable for subsequent weighted fusion and model training;

[0049] The standardized data is weighted and fused using the weighted average method to generate comprehensive indicators:

[0050]

[0051] In the above formula, F i represents the fusion value of the i-th indicator, w ij represents the weight of the jth platform, and the weight parameter w ij Based on the importance and data quality of each platform, D′ ij Represents the i-th indicator data of the j-th platform. The standardized data of each platform is integrated through the weighted average method to generate a comprehensive indicator. This method allows different weights to be assigned according to the data quality and importance of each platform, thereby reflecting the relative importance of the platforms in the comprehensive evaluation. Weighted fusion can comprehensively consider the impact of multiple data sources and improve the accuracy and representativeness of the comprehensive indicator;

[0052] Exponential smoothing is used to smooth the fused data to reduce the impact of noise:

[0053] S t =αF t +(1-α)S t-1

[0054] In the above formula, S t represents the smoothed value at time t, F tRepresents the fusion value at time t, α represents the smoothing coefficient, and exponential smoothing is used to smooth the fused data, reduce the noise and fluctuations in the data, and extract the long-term trend of the data. Smoothing can effectively reduce the impact of short-term fluctuations in the data on the analysis results, making the results more stable and reliable, which is very important in analysis and prediction, especially when dealing with seasonal or trending data. Smoothing can improve the robustness of the model and the accuracy of prediction.

[0055] S4. Use the effect evaluation model to evaluate the effect of marketing activities, and use the following formula to calculate the click-through rate (CTR), conversion rate (CVR), user retention rate (RR) and customer lifetime value (CLV) to comprehensively evaluate the effect of marketing activities. The construction of the effect evaluation model aims to comprehensively measure and evaluate the effect of marketing activities by calculating multi-dimensional indicators.

[0056] It should be noted that the specific construction method of the effect evaluation model includes: multi-dimensional indicator calculation, calculation of click-through rate (CTR), conversion rate (CVR), user retention rate (RR) and user lifetime value (CLV), and comprehensive evaluation of marketing campaign effects:

[0057]

[0058]

[0059] In the above formula, Clicks represents the number of clicks, Impressions represents the number of impressions, Conversions represents the number of conversions, Retained Users represents the number of retained users, Total Users represents the total number of users, R represents the revenue in the tth period, and C trepresents the cost of the tth period, d represents the discount rate, T represents the total number of periods in the user life cycle, the click-through rate (CTR) measures the frequency of advertisement clicks and is an important indicator for evaluating advertisement attractiveness and user engagement. A high click-through rate usually indicates that the advertisement content is attractive and can attract users to click. The conversion rate (CVR) measures the frequency of users completing the expected action (such as purchase, registration) after clicking on the advertisement and is an important indicator for evaluating advertisement effectiveness and return on investment. A high conversion rate indicates that the advertisement has a high actual effect and can effectively promote user conversion. The user retention rate (RR) measures the proportion of users who continue to use the product or service after the initial participation and is an important indicator for evaluating product or service quality and user satisfaction. A high retention rate indicates that the user's continued interest and loyalty to the product or service. The user lifetime value (CLV) measures the total value created by a user for the enterprise during his or her life cycle and is an important indicator for evaluating the long-term value of users and the profitability of the enterprise. A high CLV indicates that the user has brought higher income to the enterprise and has higher economic value.

[0060] Comprehensive score calculation, using fuzzy comprehensive evaluation to comprehensively score each evaluation indicator and quantify the effect of marketing activities:

[0061]

[0062] In the above formula, I′ k represents the normalized kth index, w k Represents the corresponding weight parameter, m represents the total number of indicators, and the fuzzy comprehensive evaluation method quantifies the overall effect of marketing activities by weighted integration of multiple indicators. This method can comprehensively consider the importance and influence of each indicator and provide a quantitative comprehensive score, which is convenient for comparing and evaluating the effects of different marketing activities. The weight parameter can be adjusted according to actual business needs and data analysis results to reflect the relative importance of each indicator in the comprehensive evaluation.

[0063] S5. Use the effect prediction model constructed based on historical data to perform effect prediction. The construction and use of the effect prediction model is to accurately predict the effects of future marketing activities.

[0064] It should be noted that the specific construction method of the effect prediction model includes: feature engineering processing. The main purpose of feature engineering is to extract and generate new features from the original data to improve the performance and effect of the model. Feature engineering can significantly improve the prediction ability of the model and the comprehensiveness of the effect evaluation. It is a key step in the machine learning process. It extracts and selects features from historical data to build an effective prediction model input:

[0065] X=[X1,X2,...,X p ]

[0066] In the above formula, X represents the feature vector, X i Represents the i-th feature, p represents the number of features. Feature engineering processing extracts and selects features from historical data to build effective prediction model inputs. By selecting and generating meaningful features, the prediction performance of the model can be improved, enabling it to better capture the laws and trends in the data, thereby improving the accuracy and stability of the prediction.

[0067] Model selection and training, select the long short-term memory network (LSTM), and optimize the model parameters through training:

[0068]

[0069] In the above formula, represents the predicted value at time t, W represents the model parameters, represents the n historical values ​​before time t, f represents the LSTM model, and the long short-term memory network (LSTM) is selected as the prediction model. The model parameters are optimized through training so that the model can learn the long-term dependencies and time series characteristics in the historical data, thereby improving the accuracy of the prediction. The LSTM model has good performance in processing time series data and can effectively capture the complex patterns and trends in the data;

[0070] Multi-step forecasting strategy uses a multi-step forecasting method to predict the marketing effects at multiple time points in the future and improve the forecast accuracy:

[0071]

[0072] In the above formula, represents the predicted value at time t+k, Represents n predicted values ​​before time t+k. The multi-step prediction strategy improves the comprehensiveness and accuracy of the prediction by iteratively predicting the effects at multiple future time points. This strategy can provide continuous predictions of marketing effects over a period of time in the future, facilitating long-term planning and decision-making.

[0073] Model evaluation, using root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE) to evaluate the effect prediction model:

[0074]

[0075] In the above formula, N represents the time step of the forecast period, and Y(t+i) are the predicted value and actual value respectively. By evaluating the error of the prediction model, the prediction performance of the model can be quantified to ensure the accuracy and reliability of its prediction results. RMSE, MAE and MAPE evaluate the prediction error from different angles respectively, providing a comprehensive model performance evaluation standard to facilitate the optimization and improvement of the model.

[0076] In a possible embodiment, firstly, data related to mobile phone marketing activities on multiple Internet platforms are collected:

[0077] Platform A: 100,000 impressions, 5,000 clicks, 100 conversions, 10,000 interactions, and 50,000 sales.

[0078] Platform B: 150,000 impressions, 6,000 clicks, 90 conversions, 8,000 interactions, and 60,000 sales.

[0079] Platform C: 120,000 impressions, 7,200 clicks, 180 conversions, 12,000 interactions, and 55,000 sales.

[0080] Assume that the sales data of platform C is missing, and use the mean imputation method to fill the missing values:

[0081]

[0082] Assuming that the number of clicks on platform A is an extreme value, the box plot method is used to identify outliers. After calculating Q1, Q3 and IQR, the upper and lower limits are determined:

[0083] Q1=5,000Q3=7,200

[0084] IQR = Q3 - Q1 = 2,200

[0085] LSL = Q1-1.5 × IQR = 1,700

[0086] USL = Q3 + 1.5 × IQR = 10,500

[0087] The number of hits is 5,000 and is within range, so there is no need to deal with outliers.

[0088] The Min-Max normalization method is used to convert the data to the [0,1] interval:

[0089]

[0090] Assuming sales range is between 50,000 and 60,000:

[0091]

[0092] Standardize the data for each platform:

[0093]

[0094] Assume that the mean number of clicks on platform A is 5,400 and the standard deviation is 1,100:

[0095]

[0096] The standardized data is weighted and fused using the weighted average method to generate comprehensive indicators:

[0097]

[0098] Assume that the weights of platforms A, B, and C are 0.4, 0.3, and 0.3 respectively:

[0099] F 点击量 =0.4×-0.364+0.3×0.545+0.3×0.927=0.314

[0100] Use exponential smoothing method to smooth the fused data:

[0101] S t =αF t +(1-α)S t-1

[0102] Assume the initial smoothing value is 0.3 and the smoothing coefficient is 0.5:

[0103] S1=0.5×0.314+0.5×0.3=0.307

[0104] Calculate multi-dimensional indicators through the effect evaluation model:

[0105]

[0106] Assume a 10% discount rate, $10,000 in revenue per period, and $2,000 in costs per period:

[0107]

[0108] Comprehensive score calculation:

[0109] CS=0.25×5+0.25×2+0.25×90+0.25×30,184.23=7,570.56

[0110] The effect prediction model performs the following tasks:

[0111] Feature Engineering Processing:

[0112] X={5,000,100,90%,30,184.23}

[0113] Model selection and training, select LSTM, optimize model parameters, multi-step prediction strategy and model evaluation through training. Assuming that the actual click rate is 4.8%, then:

[0114]

[0115] Through the above specific embodiments, the present invention can effectively collect and process multi-platform data, build accurate evaluation and prediction models, and provide strong support for the optimization and decision-making of marketing activities.

[0116] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0117] (1) In the present invention, the marketing campaign effect is comprehensively evaluated through multi-dimensional indicator calculation and comprehensive score calculation. Key performance indicators such as click-through rate, conversion rate, user retention rate and user lifetime value are used, combined with fuzzy comprehensive evaluation method, so that the evaluation results are more accurate and comprehensive, which is helpful for the optimization and decision-making of marketing strategies;

[0118] (2) In the present invention, a long short-term memory network (LSTM) model is used to improve the performance of the effect prediction model through feature engineering processing, model selection and training, multi-step prediction strategy and other steps. Through deep learning and analysis of historical data, future marketing effects can be accurately predicted, helping enterprises to carry out effective market planning and resource allocation;

[0119] (3) In the present invention, during the data processing process, the data quality and consistency are ensured by removing missing values ​​and outliers, adopting the Min-Max normalization method, standardization processing and weighted fusion, and using the exponential smoothing method to reduce data noise, making the data fusion model more robust and reliable, providing high-quality data support for subsequent evaluation and prediction.

[0120] The above contents are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0121] There are a few points to note:

[0122] (1) The drawings of the embodiments of the present invention only relate to the structures related to the embodiments of the present invention, and other structures may refer to the general design.

[0123] (2) For the sake of clarity, in the drawings used to describe the embodiments of the present invention, the thickness of the layers or regions is exaggerated or reduced, that is, these drawings are not drawn according to the actual scale. It is understood that when an element such as a layer, film, region or substrate is referred to as being "on" or "under" another element, the element may be "directly" "on" or "under" the other element or there may be intermediate elements.

[0124] (3) In the absence of conflict, the embodiments of the present invention and the features therein may be combined with each other to obtain new embodiments.

[0125] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for constructing a multi-platform Internet marketing campaign effect evaluation and prediction model, characterized in that: include: S1. Collect marketing activity related data from multiple Internet platforms, including exposure, click-through rate, conversion rate, interaction volume and sales data; S2. Clean, normalize and feature engineer the collected data to generate a unified data format for subsequent analysis; S3. Based on machine learning algorithms, build multi-level evaluation and prediction models, including data fusion models, effect evaluation models, and effect prediction models; S4. Evaluate the effect of the marketing campaign using the effect evaluation model, and use the following formula to calculate the click-through rate (CTR), conversion rate (CVR), user retention rate (RR) and customer lifetime value (CLV) to comprehensively evaluate the effect of the marketing campaign; S5. Use the effect prediction model constructed based on historical data to perform effect prediction.

2. A method for constructing a multi-platform Internet marketing activity effect evaluation and prediction model according to claim 1, characterized in that: The S2 specifically includes: Missing values ​​and outliers were removed, and the data were transformed to the [0,1] interval using the Min-Max normalization method.

3. A method for constructing a multi-platform Internet marketing activity effect evaluation and prediction model according to claim 2, characterized in that: The removal of missing values ​​specifically includes: For numerical data, the following methods are used to fill missing values: In the above formula, the upper and lower terms are respectively i When the value is non-missing and when x i When x is a missing value, i represents the i-th data point, and n represents the number of non-missing data points.

4. A method for constructing a multi-platform Internet marketing activity effect evaluation and prediction model according to claim 2, characterized in that: The removal of outliers specifically includes: The box plot method was used to identify outliers, and the interquartile range (IQR) was calculated to determine the range of outliers: IQR=Q3-Q1 LSL=Q1-1.5×IQR USL=Q3+1.5×IQR In the above formula, Q1 and Q3 represent the first quartile and the third quartile respectively, LSL represents the lower limit, and USL represents the upper limit. Data exceeding the upper and lower limits are regarded as outliers and are processed by the following method:

5. A method for constructing a multi-platform Internet marketing activity effect evaluation and prediction model according to claim 2, characterized in that: The normalization specifically includes: The processed data is normalized by Min-Max and converted to the interval [0,1]. The formula is as follows: In the above formula, x i represents the original data, x′ i represents the normalized data, x min and x max 35 represent the minimum and maximum values ​​of the data respectively.

6. A method for constructing a multi-platform Internet marketing activity effect evaluation and prediction model according to claim 1, characterized in that: The data fusion model in S3 specifically includes: The data of each platform is standardized to be in the same dimension using the following formula: In the above formula, D′ ij represents the standardized data, D ij represents the original data, μ j and σ j Respectively represent the mean and standard deviation of the j-th platform; The standardized data is weighted and fused using the weighted average method to generate comprehensive indicators: In the above formula, F i represents the fusion value of the i-th indicator, w ij represents the weight of the jth platform, and the weight parameter w ij Based on the importance and data quality of each platform, D′ ij represents the i-th indicator data of the j-th platform, D ij is the weight parameter; Exponential smoothing is used to smooth the fused data to reduce the impact of noise: S t =αF t +(1-α)S t-1 In the above formula, S t represents the smoothed value at time t, F t represents the fusion value at time t, and α represents the smoothing coefficient.

7. A method for constructing a multi-platform Internet marketing activity effect evaluation and prediction model according to claim 1, characterized in that: The effect evaluation model in S4 specifically includes: Multi-dimensional indicator calculation, including click-through rate (CTR), conversion rate (CVR), user retention rate (RR) and customer lifetime value (CLV), comprehensively evaluates the effectiveness of marketing activities: In the above formula, Clicks represents the number of clicks, Impressions represents the number of impressions, Conversions represents the number of conversions, Retained Users represents the number of retained users, Total Users represents the total number of users, R represents the revenue in the tth period, and C t represents the cost of period t, d represents the discount rate, and T represents the total number of periods in the user's life cycle; Comprehensive score calculation, using fuzzy comprehensive evaluation to comprehensively score each evaluation indicator and quantify the effect of marketing activities: In the above formula, I' k represents the normalized kth index, w k represents the corresponding weight parameter, and m represents the total number of indicators.

8. The method for constructing a multi-platform Internet marketing activity effect evaluation and prediction model according to claim 1, characterized in that: The effect prediction model in S5 includes: Feature engineering processing, feature extraction and selection of historical data, and construction of effective prediction model input: X=[X1,X2,...,X p ] In the above formula, X represents the feature vector, X i represents the i-th feature, and p represents the number of features; Model selection and training, select the long short-term memory network (LSTM), and optimize the model parameters through training: In the above formula, represents the predicted value at time t, W represents the model parameters, represents n historical values ​​before time t, and f represents the LSTM model; Multi-step forecasting strategy uses a multi-step forecasting method to predict the marketing effects at multiple time points in the future and improve the forecast accuracy: In the above formula, represents the predicted value at time t+k, Represents n predicted values ​​before time t+k; Model evaluation, using root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE) to evaluate the effect prediction model: In the above formula, N represents the time step of the forecast period, and Y(t+i) are the predicted value and actual value respectively.

Images