Marketing model data processing method, system and computer readable storage medium
By constructing a marketing model data processing method, combined with a multi-dimensional interpretation system and feedback mechanism, the problem of insufficient interpretability of marketing models in banking business scenarios was solved, marketing efficiency and accuracy were improved, and the accuracy of the model was gradually increased.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA MERCHANTS BANK
- Filing Date
- 2023-01-06
- Publication Date
- 2026-05-15
AI Technical Summary
Existing marketing models lack an effective explanatory framework in banking scenarios, resulting in marketers having only a superficial understanding of the model's results, which reduces model usage and marketing effectiveness.
This paper proposes a marketing model data processing method. By acquiring marketing information and performing data mining, the method makes predictions based on a pre-set model and interprets the prediction results using a multi-dimensional model result interpretation system. This method showcases prominent marketing categories under different dimensions, helping business personnel understand and determine marketing entry points.
It improves the interpretability of the marketing model, helps business personnel accurately find marketing entry points, improves marketing efficiency and accuracy, and improves the accuracy of the model by correcting the model through feedback results.
Smart Images
Figure CN116091113B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of marketing scenarios, and in particular to a marketing model data processing method, system, and computer-readable storage medium. Background Technology
[0002] In banking scenarios, marketing models are typically built to assist marketing personnel in their work. However, existing model feedback systems often only provide rigid, theoretical results. In actual banking marketing situations, some marketing personnel are unfamiliar with these models and lack professional knowledge about them. Therefore, when faced with technical model results provided by the middle and back offices, they often have a vague understanding and are hesitant to use them, leading to a decrease in the utilization rate of marketing models and a reduction in marketing effectiveness.
[0003] The root cause of this phenomenon lies in the lack of an interpretable system for the model's prediction results. Therefore, it is necessary to propose an interpretable system for marketing model prediction results to bridge the technical gap between the complex mathematical principles behind the model and business personnel who are not yet familiar with the model.
[0004] Current mainstream approaches typically involve providing the model's key features to explain the results holistically, but this fails to provide interpretation from an individual customer's perspective. While scorecard models can achieve individualized interpretability, they rely on a limited number of features and cannot support the use of large amounts of feature data to build more complex models. These existing methods all result in poor interpretability of marketing models, hindering improvements in marketing efficiency and accuracy. Summary of the Invention
[0005] The main objective of this invention is to provide a marketing model data processing method, system, and computer-readable storage medium, which aims to improve the interpretability of marketing models, assist marketers in accurately identifying marketing entry points, determining marketing methods, and improving marketing efficiency and accuracy.
[0006] To achieve the above objectives, the present invention provides a marketing model data processing method, the method comprising the following steps:
[0007] Obtain marketing information;
[0008] Data mining is performed on the marketing targets based on the marketing information to obtain multi-dimensional category data reflecting the marketing targets.
[0009] Based on a preset marketing model, the multi-dimensional category data of marketing is predicted to obtain the model prediction results in the marketing scenario.
[0010] The model prediction results are interpreted based on a pre-built model result interpretation system to obtain and display prominent marketing categories under different dimensions.
[0011] Optionally, the method further includes:
[0012] Obtain feedback from marketing personnel;
[0013] The model's predictions were revised based on feedback from the marketers.
[0014] Optionally, before the step of interpreting the model prediction results based on a pre-built model result interpretation system to obtain and display prominent marketing categories under different dimensions, the method further includes:
[0015] Acquire marketing scenario data;
[0016] The marketing scenario data is anonymized.
[0017] Based on the de-identified marketing scenario data, a profile system is constructed through multiple dimensions to intuitively interpret the model prediction results, resulting in a model result interpretation system. Each dimension includes several subcategories.
[0018] Optionally, the step of constructing a profile system based on the marketing scenario data to intuitively interpret the model prediction results, thereby obtaining the model result interpretation system, includes:
[0019] Based on the aforementioned marketing scenario data, a marketing profile combining dynamic and static elements is constructed through multiple dimensions;
[0020] Based on the aforementioned marketing profile, a model result interpretation system is obtained to provide an intuitive interpretation of the model's prediction results.
[0021] Optionally, the step of constructing a marketing profile combining dynamic and static elements based on the marketing scenario data through multiple dimensions includes:
[0022] Based on the marketing scenario data, obtain static user characteristics from multiple dimensions, and construct static profiles based on the static user characteristics;
[0023] Based on the marketing scenario data, dynamic user characteristics in multiple dimensions are obtained. Based on these dynamic user characteristics, a dynamic profile is constructed through global feature comparison and intra-dimensional feature comparison.
[0024] A marketing profile is constructed by combining the static and dynamic profiles.
[0025] Optionally, the step of obtaining dynamic user characteristics across multiple dimensions based on the marketing scenario data, and constructing a dynamic profile based on these dynamic user characteristics through global feature comparison and intra-dimensional feature comparison, includes:
[0026] Based on the marketing scenario data, dynamic user characteristics of different categories in multiple dimensions are obtained;
[0027] For each dynamic user feature, perform a global vertical comparison of the data, calculate the ranking of a single data point in the global data within the current category, and obtain the percentage ranking of a single data point within the current category;
[0028] The top-ranked data points are used to create a dynamic global profile.
[0029] Based on the percentage of data obtained from global vertical comparison, horizontal comparison is performed within the dimensions of the data to construct a dynamic profile.
[0030] Optionally, the step of constructing a dynamic profile by performing intra-dimensional horizontal comparison of data based on the percentage of data obtained from global vertical comparison includes:
[0031] Based on the percentage of data obtained from global vertical comparison, the categories within each dimension of each data point are filtered to remove invalid, abnormal, and empty categories;
[0032] After filtering each data point, we perform a horizontal comparison of each category within each dimension, and then display the top-performing categories to obtain a unique dynamic profile for each data point.
[0033] Optionally, the subcategories are divided into numerical and character types. The step of obtaining dynamic user characteristics across multiple dimensions based on the marketing scenario data, and constructing a dynamic profile based on these dynamic user characteristics through global feature comparison and intra-dimensional feature comparison, further includes:
[0034] For the numerical categories in the marketing scenario data, calculate the global ranking percentage of the marketing object;
[0035] For the character-type categories in the marketing scenario data, the character-type categories are transformed, and each character-type category is assigned a corresponding weight through statistical methods. The ranking percentage of the sub-category is calculated based on the weight.
[0036] Optionally, the step of obtaining a model result interpretation system based on the marketing profile to intuitively interpret the model prediction results includes:
[0037] Based on the aforementioned marketing profile, a communication script system is constructed to process the displayed categories using professional communication techniques.
[0038] Optionally, the step of obtaining a model result interpretation system based on the marketing profile to intuitively interpret the model prediction results further includes:
[0039] Establish a global category ranking relationship table, a dimension-based category ranking relationship table, a category mapping control relationship table, a sales script conversion processing table, a model prediction result correction table, and a marketing personnel actual result feedback correction table, so as to construct the model result interpretation system through the interrelationship between the tables.
[0040] Optionally, before the step of predicting the multi-dimensional marketing category data based on a preset marketing model to obtain the model prediction results in the marketing scenario, the method further includes:
[0041] The marketing model is constructed by including:
[0042] Build the initial dataset;
[0043] The model is trained and validated using the initial dataset to obtain a trained initial model.
[0044] The trained initial model is used to predict real-world scene data to obtain the initial model prediction results;
[0045] The initial model prediction results are compared with the actual results, and the final initial model prediction results are determined based on the comparison results.
[0046] Based on the final initial model prediction results, the data with incorrect model predictions are stratified and sampled as a supplementary training dataset for the next round of model iteration.
[0047] Based on the final initial model prediction results, the data that was correctly predicted by the model is stratified and used as the training dataset for the next round of model iteration.
[0048] The training dataset and the training supplementary dataset are used as the data basis for the next iteration of the model. The model is iterated until the expected stopping condition is met, and the model iteration ends, resulting in a trained marketing model.
[0049] This invention also proposes a marketing model data processing system, the system comprising:
[0050] The acquisition module is used to acquire marketing information;
[0051] The data mining module is used to perform data mining on the marketing object based on the marketing information to obtain multi-dimensional category data reflecting the marketing object;
[0052] The prediction module is used to predict the multi-dimensional category data of marketing based on a preset marketing model, and obtain the model prediction results in the marketing scenario.
[0053] The explanation and display module is used to interpret the model prediction results based on the pre-built model result interpretation system, and to obtain and display the prominent marketing categories under different dimensions.
[0054] This invention also proposes a marketing model data processing system, which includes a memory, a processor, and a marketing model data processing program stored in the memory and executable on the processor. When the marketing model data processing program is executed by the processor, it implements the marketing model data processing method described above.
[0055] This invention also proposes a computer-readable storage medium storing a marketing model data processing program, which, when executed by a processor, implements the marketing model data processing method described above.
[0056] This invention proposes a marketing model data processing method, system, and computer-readable storage medium. The method involves acquiring marketing information; performing data mining on the marketing object based on the marketing information to obtain multi-dimensional marketing category data reflecting the marketing object; predicting the multi-dimensional marketing category data based on a preset marketing model to obtain model prediction results for the marketing scenario; and interpreting the model prediction results based on a pre-built model result interpretation system to obtain and display prominent marketing categories under different dimensions. This invention focuses on addressing the confusion faced by business personnel when using model results. By combining specific business scenarios, it constructs a complete and highly interpretable model result interpretation system, and intuitively and dynamically displays prominent marketing categories under different dimensions on a front-end page. This allows business personnel to combine model prediction results with their own experience to make targeted marketing decisions. This embodiment bridges the technical gap between the complex mathematical principles behind the model and business personnel unfamiliar with the model, helping them accurately find marketing entry points and determine marketing methods in specific usage scenarios through multiple dimensions such as model prediction results, their own experience, and the model result interpretation system, thereby improving marketing efficiency, accuracy, and the lower limit of marketing effectiveness. Furthermore, the model prediction results can be corrected based on actual marketing scenarios and feedback from marketers. Incorrectly predicted samples can be incorporated into the next update and iteration of the model, thereby gradually improving the accuracy of the marketing model. Attached Figure Description
[0057] Figure 1 This is a schematic diagram of the functional modules of the terminal device to which the marketing model data processing device of this invention belongs;
[0058] Figure 2This is a flowchart illustrating the first embodiment of the marketing model data processing method of the present invention;
[0059] Figure 3 This is a flowchart illustrating the second embodiment of the marketing model data processing method of the present invention;
[0060] Figure 4 This is a flowchart illustrating the third embodiment of the marketing model data processing method of the present invention.
[0061] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0062] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
[0063] The main solution of this invention is as follows: First, obtain marketing information. Second, perform data mining on the marketing target based on the marketing information to obtain multi-dimensional marketing category data reflecting the marketing target. Third, predict the multi-dimensional marketing category data based on a preset marketing model to obtain model prediction results for the marketing scenario. Fourth, interpret the model prediction results based on a pre-built model result interpretation system to obtain and display prominent marketing categories under different dimensions. This invention focuses on solving the confusion faced by business personnel when using model results. By combining specific business scenarios, it constructs a complete and highly interpretable model result interpretation system. The front-end page intuitively and dynamically displays prominent marketing categories under different dimensions, enabling business personnel to combine model prediction results with their own experience to make targeted marketing decisions. This embodiment bridges the technical gap between the complex mathematical principles behind the model and business personnel unfamiliar with the model. It helps business personnel accurately find marketing entry points and determine marketing methods in specific usage scenarios through multiple dimensions, including model prediction results, their own experience, and the model result interpretation system, thereby improving marketing efficiency, accuracy, and the lower limit of marketing effectiveness. Furthermore, the model prediction results can be corrected based on actual marketing scenarios and feedback from marketers. Incorrectly predicted samples can be incorporated into the next update and iteration of the model, thereby gradually improving the accuracy of the marketing model.
[0064] This invention addresses the issue that existing model result feedback systems often only provide rigid model results. In actual banking marketing scenarios, some marketing personnel are unfamiliar with the models and lack professional knowledge about them. Therefore, when faced with technical model results provided by the back office, they often have only a superficial understanding and are hesitant to use them, leading to a decrease in the usage rate of marketing models and a reduction in marketing effectiveness. The root cause of this phenomenon lies in the lack of an interpretable system for model prediction results. Current mainstream approaches provide the model's key features to explain the results holistically, but fail to interpret them from the perspective of individual clients. Alternatively, they may use scorecard models to achieve interpretability at the individual client level, but scorecard models only use a limited number of features and cannot utilize large amounts of feature data to build more complex models. All existing methods result in poor interpretability of marketing models, hindering the improvement of marketing efficiency and accuracy.
[0065] Based on this, embodiments of the present invention provide a solution that, by combining specific business scenarios, constructs a complete and highly interpretable model result interpretation system. The system intuitively and dynamically displays prominent marketing categories under different dimensions through a front-end page, enabling business personnel to combine the model prediction results with their own experience to make targeted marketing decisions. This can dispel the confusion of marketing personnel, improve the interpretability of the marketing model, assist marketing personnel in accurately finding marketing entry points, determining marketing methods, and improving marketing efficiency and accuracy.
[0066] Specifically, refer to Figure 1 , Figure 1 This is a schematic diagram of the functional modules of the terminal device to which the marketing model data processing device of the present invention belongs. This marketing model data processing device can be a device independent of the terminal device, and it can be carried on the terminal device or system in the form of hardware or software. The terminal device can be a smart mobile terminal such as a mobile phone or tablet computer, or it can be a network device such as a server.
[0067] In this embodiment, the terminal device to which the marketing model data processing device belongs includes at least an output module 110, a processor 120, a memory 130, and a communication module 140.
[0068] The memory 130 stores the operating system and marketing model data processing program; the output module 110 can be a display screen, speaker, etc. The communication module 140 can include a WIFI module, a mobile communication module, and a Bluetooth module, etc., and communicates with external devices or servers through the communication module 140.
[0069] In one embodiment, when the marketing model data processing program in memory 130 is executed by the processor, it performs the following steps:
[0070] Obtain marketing information;
[0071] Data mining is performed on the marketing targets based on the aforementioned marketing information to obtain multi-dimensional marketing category data reflecting the marketing targets.
[0072] Based on a preset marketing model, the multi-dimensional category data of marketing is predicted to obtain the model prediction results in the marketing scenario.
[0073] The model prediction results are interpreted based on a pre-built model result interpretation system to obtain and display prominent marketing categories under different dimensions.
[0074] Furthermore, when the marketing model data processing program in memory 130 is executed by the processor, it also performs the following steps:
[0075] Obtain feedback from marketing personnel;
[0076] The model's predictions were revised based on feedback from the marketers.
[0077] Furthermore, when the marketing model data processing program in memory 130 is executed by the processor, it also performs the following steps:
[0078] Acquire marketing scenario data;
[0079] Based on the marketing scenario data, a profile system is constructed through multiple dimensions to intuitively interpret the model prediction results, resulting in a model result interpretation system. Each dimension includes several subcategories.
[0080] Furthermore, when the marketing model data processing program in memory 130 is executed by the processor, it also performs the following steps:
[0081] The marketing scenario data is anonymized.
[0082] Furthermore, when the marketing model data processing program in memory 130 is executed by the processor, it also performs the following steps:
[0083] Based on the aforementioned marketing scenario data, a marketing profile combining dynamic and static elements is constructed through multiple dimensions;
[0084] Based on the aforementioned marketing profile, a model result interpretation system is obtained to provide an intuitive interpretation of the model's prediction results.
[0085] Furthermore, when the marketing model data processing program in memory 130 is executed by the processor, it also performs the following steps:
[0086] Based on the marketing scenario data, obtain static user characteristics from multiple dimensions, and construct static profiles based on the static user characteristics;
[0087] Based on the marketing scenario data, dynamic user characteristics in multiple dimensions are obtained. Based on these dynamic user characteristics, a dynamic profile is constructed through global feature comparison and intra-dimensional feature comparison.
[0088] A marketing profile is constructed by combining the static and dynamic profiles.
[0089] Furthermore, when the marketing model data processing program in memory 130 is executed by the processor, it also performs the following steps:
[0090] Based on the marketing scenario data, dynamic user characteristics of different categories in multiple dimensions are obtained;
[0091] For each dynamic user feature, perform a global vertical comparison of the data, calculate the ranking of a single data point in the global data within the current category, and obtain the percentage ranking of a single data point within the current category;
[0092] The top-ranked data points are used to create a dynamic global profile.
[0093] Based on the percentage of data obtained from global vertical comparison, horizontal comparison is performed within the dimensions of the data to construct a dynamic profile.
[0094] Furthermore, when the marketing model data processing program in memory 130 is executed by the processor, it also performs the following steps:
[0095] Based on the percentage of data obtained from global vertical comparison, the categories within each dimension of each data point are filtered to remove invalid, abnormal, and empty categories;
[0096] After filtering each data point, we perform a horizontal comparison of each category within each dimension, and then display the top-performing categories to obtain a unique dynamic profile for each data point.
[0097] Furthermore, when the marketing model data processing program in memory 130 is executed by the processor, it also performs the following steps:
[0098] For the numerical categories in the marketing scenario data, calculate the global ranking percentage of the marketing object;
[0099] For the character-type categories in the marketing scenario data, the character-type categories are transformed, and each character-type category is assigned a corresponding weight through statistical methods. The ranking percentage of the sub-category is calculated based on the weight.
[0100] Furthermore, when the marketing model data processing program in memory 130 is executed by the processor, it also performs the following steps:
[0101] Based on the aforementioned marketing profile, a communication script system is constructed to process the displayed categories using professional communication techniques.
[0102] Furthermore, when the marketing model data processing program in memory 130 is executed by the processor, it also performs the following steps:
[0103] Establish a global category ranking relationship table, a dimension-based category ranking relationship table, a category mapping control relationship table, a sales script conversion processing table, a model prediction result correction table, and a marketing personnel actual result feedback correction table, so as to construct the model result interpretation system through the interrelationship between the tables.
[0104] Furthermore, when the marketing model data processing program in memory 130 is executed by the processor, it also performs the following steps:
[0105] The marketing model is constructed by including:
[0106] Build the initial dataset;
[0107] The model is trained and validated using the initial dataset to obtain a trained initial model.
[0108] The trained initial model is used to predict real-world scene data to obtain the initial model prediction results;
[0109] The initial model prediction results are compared with the actual results, and the final initial model prediction results are determined based on the comparison results.
[0110] Based on the final initial model prediction results, the data with incorrect model predictions are stratified and sampled as a supplementary training dataset for the next round of model iteration.
[0111] Based on the final initial model prediction results, the data that was correctly predicted by the model is stratified and used as the training dataset for the next round of model iteration.
[0112] The training dataset and the training supplementary dataset are used as the data basis for the next iteration of the model. The model is iterated until the expected stopping condition is met, and the model iteration ends, resulting in a trained marketing model.
[0113] This embodiment, through the above-described scheme, specifically involves acquiring marketing information; performing data mining on the marketing target based on the marketing information to obtain multi-dimensional marketing category data reflecting the marketing target; predicting the multi-dimensional marketing category data based on a preset marketing model to obtain model prediction results under the marketing scenario; and interpreting the model prediction results based on a pre-built model result interpretation system to obtain and display prominent marketing categories under different dimensions. This invention focuses on solving the confusion faced by business personnel when using model results. By combining specific business scenarios, it constructs a complete and highly interpretable model result interpretation system, and intuitively and dynamically displays prominent marketing categories under different dimensions on the front-end page, enabling business personnel to combine model prediction results with their own experience to make targeted marketing decisions. This embodiment can bridge the technical gap between the complex mathematical principles behind the model and business personnel unfamiliar with the model, helping them accurately find marketing entry points and determine marketing methods in specific usage scenarios through multiple dimensions such as model prediction results, their own experience, and the model result interpretation system, thereby improving marketing efficiency, accuracy, and the lower limit of marketing effectiveness. Furthermore, the model prediction results can be corrected based on actual marketing scenarios and feedback from marketers. Incorrectly predicted samples can be incorporated into the next update and iteration of the model, thereby gradually improving the accuracy of the marketing model.
[0114] Based on, but not limited to, the terminal device architecture described above, embodiments of the method of the present invention are proposed.
[0115] Reference Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of the marketing model data processing method of the present invention.
[0116] like Figure 2 As shown in the figure, an embodiment of the present invention proposes a marketing model data processing method, the method comprising the following steps:
[0117] Step S101: Obtain marketing information;
[0118] The method in this embodiment is mainly applied to marketing scenarios such as banking business, but it can also be applied to marketing scenarios of other businesses. This embodiment uses the marketing scenario of banking business as an example, but it does not constitute a limitation.
[0119] First, obtain actual marketing information, which may include behavioral data of each marketing target (i.e., customer), relevant fixed information (such as basic attribute information such as age, gender, and education level of the marketing target), income, payroll status, credit status, etc.
[0120] Step S102: Perform data mining on the marketing object based on the marketing information to obtain multi-dimensional marketing category data reflecting the marketing object;
[0121] Data mining is performed on specific marketing targets to obtain multi-dimensional category data that can effectively reflect the marketing targets.
[0122] The dimensions of multi-dimensional category data can include: basic attribute information of the marketing target such as age, gender, and education level, income, payroll status, and credit status.
[0123] Step S103: Based on the preset marketing model, predict the multi-dimensional category data of marketing to obtain the model prediction results in the marketing scenario;
[0124] In this embodiment, a marketing model is pre-built based on a sample dataset. The marketing model can be used to predict the input marketing data and obtain the model prediction results in the marketing scenario.
[0125] Step S104: Based on the pre-built model result interpretation system, the model prediction results are interpreted to obtain and display the prominent marketing categories under different dimensions.
[0126] Because existing model result feedback systems often only provide rigid model results, and in actual banking marketing scenarios, some marketing personnel are not familiar with the models and lack professional knowledge of the models, they often have a vague understanding and doubt when faced with the technical model results provided by the middle and back offices. This leads to a situation where they want to use the models but dare not, thus reducing the usage rate of marketing models and the marketing effectiveness of marketing personnel.
[0127] This embodiment introduces a model result interpretation system to explain the model's prediction results, revealing prominent marketing categories across different dimensions and presenting them to marketers. This bridges the technical gap between the complex mathematical principles behind the model and business personnel unfamiliar with it. It helps business personnel accurately identify marketing entry points and determine marketing methods in specific use cases by leveraging model prediction results, their own experience, and the model result interpretation system, thereby improving marketing efficiency, accuracy, and the lower limit of marketing effectiveness.
[0128] This embodiment, through the above-described scheme, specifically involves: acquiring marketing information; performing data mining on the marketing target based on the marketing information to obtain multi-dimensional marketing category data reflecting the marketing target; predicting the multi-dimensional marketing category data based on a preset marketing model to obtain model prediction results under the marketing scenario; and interpreting the model prediction results based on a pre-built model result interpretation system to obtain and display prominent marketing categories under different dimensions. This invention focuses on solving the confusion faced by business personnel when using model results. By combining specific business scenarios, it constructs a complete and highly interpretable model result interpretation system, and intuitively and dynamically displays prominent marketing categories under different dimensions on the front-end page, enabling business personnel to combine model prediction results with their own experience to make targeted judgments and carry out marketing activities.
[0129] Furthermore, the method also includes: obtaining feedback from marketers; and correcting the model prediction results based on the feedback from marketers, thereby providing more accurate marketing results on the one hand, and incorporating incorrect prediction samples into the next update iteration of the model on the other hand, so as to gradually improve the accuracy of the model.
[0130] Reference Figure 3 , Figure 3 This is a flowchart illustrating the second embodiment of the marketing model data processing method of the present invention.
[0131] like Figure 3 As shown, the embodiments of the present invention are based on the above. Figure 2 In the embodiment shown, in step S104 above, the model prediction results are interpreted based on a pre-built model result interpretation system to obtain and display prominent marketing categories under different dimensions (before...). Figure 3 (Taking an example before step S101) also includes:
[0132] Step S1002: Obtain marketing scenario data;
[0133] Step S1003: Based on the marketing scenario data, construct a profile system through multiple dimensions to intuitively interpret the model prediction results, thereby obtaining the model result interpretation system, wherein each dimension includes several subcategories.
[0134] Compared to the above Figure 2 The embodiment shown also includes a technical solution for constructing a model result interpretation system.
[0135] Specifically, as one implementation method, the first step is to acquire data from various marketing scenarios.
[0136] Marketing scenario data can be obtained from various online or business platforms. As the foundational data for building the interpretation system of model results, marketing scenario data can include: behavioral data of various marketing targets and relevant fixed information (such as the age, gender, and education level of the marketing targets).
[0137] One approach is to anonymize the data obtained from various marketing scenarios after acquisition, thereby improving data security and the accuracy of subsequent data processing.
[0138] Then, based on the marketing scenario data, a profile system is constructed from multiple dimensions to intuitively interpret the model prediction results, thus obtaining the model result interpretation system.
[0139] For real-world marketing scenarios, regarding the interpretability of results, after anonymizing the marketing scenario data, user profiles are constructed across multiple dimensions (the number of dimensions can be set as needed, such as 5 dimensions). Each dimension has several subcategories (such as 40), thus building a two-tier model result interpretability system. These subcategories include various user characteristics.
[0140] For example, taking 5 dimensions including 40 subcategories, the 5 dimensions can include: user's age, gender, education level, income, and payroll status.
[0141] The approximately 40 subcategories are divided into two types: numerical and character-based. For numerical categories, the overall ranking percentage of the target audience is calculated; the higher the percentage, the more significant the advantage of the target audience within that category. For character-based categories, conversion is analyzed, and statistical methods are used to assign corresponding weights to each value. Based on these weights, the ranking percentage for each subcategory is calculated.
[0142] Then, for the five categories, the percentage of subcategories within each dimension is dynamically ranked according to the actual situation of the marketing target. The higher the score, the higher the ranking. This is called the category importance ranking within the dimension. Based on the ranking results, the categories are dynamically displayed on the front-end page.
[0143] The dynamic category display is designed to take into account that user behavior changes daily, and the corresponding ranking results also change. Therefore, the display is dynamically based on the ranking results.
[0144] Specifically, as one implementation method, the steps of constructing a profile system for intuitively interpreting model prediction results through multiple dimensions to obtain a model result interpretation system may include:
[0145] Based on the aforementioned marketing scenario data, a marketing profile combining dynamic and static elements is constructed through multiple dimensions;
[0146] Based on the aforementioned marketing profile, a model result interpretation system is obtained to provide an intuitive interpretation of the model's prediction results.
[0147] The step of constructing a marketing profile combining dynamic and static elements based on the marketing scenario data through multiple dimensions may include:
[0148] Based on the marketing scenario data, obtain static user characteristics from multiple dimensions, and construct static profiles based on the static user characteristics;
[0149] Among them, static profiles mainly consist of fixed information that will not change over a relatively long period of time, which is scattered across the above-mentioned dimensions.
[0150] Based on the marketing scenario data, dynamic user characteristics in multiple dimensions are obtained. Based on these dynamic user characteristics, a dynamic profile is constructed through global feature comparison and intra-dimensional feature comparison.
[0151] The dynamic profile primarily comprises information that changes over time and is prone to change, with its features also distributed across the aforementioned dimensions. Within the interpretability framework, the construction of the dynamic profile is determined by selecting a percentage ranking.
[0152] Then, marketing profiles are constructed by combining static and dynamic profiles.
[0153] Furthermore, dynamic profiles can be constructed through two methods: global feature comparison and intra-dimensional feature comparison. The specific implementation is as follows:
[0154] First, based on the marketing scenario data, dynamic user characteristics of different categories across multiple dimensions are obtained;
[0155] For each dynamic user feature, perform a global vertical comparison of the data, calculate the ranking of a single data point in the global data within the current category, and obtain the percentage ranking of a single data point within the current category;
[0156] The top-ranked data points are used to create a dynamic global profile.
[0157] Based on the percentage data obtained from global vertical comparison, horizontal comparisons within the data dimensions are performed to construct a dynamic profile. That is, the percentage ranking obtained from global vertical comparison further serves as the data basis for horizontal comparisons within the dimensions.
[0158] The horizontal comparison within a data dimension is calculated using the percentage of data obtained from the global vertical comparison. Specific steps include:
[0159] Based on the percentage of data obtained from global vertical comparison, the categories within each dimension of each data point are filtered to remove invalid, abnormal, and empty categories;
[0160] For each piece of data, after filtering, the categories within each dimension are compared horizontally, and the top few categories by percentage are displayed to obtain different dynamic profiles for each data point.
[0161] For example:
[0162] If a customer has 20 subcategories (i.e. 20 user characteristics), classify these 20 subcategories, for example, into five categories, i.e., five dimensions.
[0163] Global vertical comparison refers to comparing the scores of each of the customer's 20 subcategories, i.e., comparing the 20 subcategories to obtain the ranking percentage;
[0164] Horizontal comparison within a dimension refers to classifying the 20 subcategories, for example, into five categories, i.e., five dimensions, and comparing the subcategories within each dimension.
[0165] This embodiment, through the above-described scheme, specifically acquires marketing information by constructing a model result interpretation system; performs data mining on the marketing object based on the marketing information to obtain multi-dimensional category data reflecting the marketing object; predicts the multi-dimensional category data based on a preset marketing model to obtain model prediction results under the marketing scenario; and interprets the model prediction results based on the pre-constructed model result interpretation system to obtain and display prominent marketing categories under different dimensions. This invention focuses on solving the confusion faced by business personnel when using model results. By combining specific business scenarios, it constructs a complete and highly interpretable model result interpretation system, and intuitively and dynamically displays prominent marketing categories under different dimensions on the front-end page, enabling business personnel to combine model prediction results with their own experience to make targeted marketing decisions. This embodiment can bridge the technical gap between the complex mathematical principles behind the model and business personnel unfamiliar with the model, helping them accurately find marketing entry points and determine marketing methods in specific usage scenarios through multiple dimensions such as model prediction results, their own experience, and the model result interpretation system, thereby improving marketing efficiency, accuracy, and the lower limit of marketing effectiveness. Furthermore, the model prediction results can be corrected based on actual marketing scenarios and feedback from marketers. Incorrectly predicted samples can be incorporated into the next update and iteration of the model, thereby gradually improving the accuracy of the marketing model.
[0166] Furthermore, as one implementation method, a sales script system can be built based on marketing profiles to convert displayed categories using professional sales scripts. That is, a sales script system is built based on static and dynamic profiles that interpret the results of model data mining, and displayed categories are converted using professional sales scripts.
[0167] Furthermore, the step of obtaining a model result interpretation system based on the marketing profile to intuitively interpret the model prediction results may also include the following schemes:
[0168] Establish a global category ranking relationship table, a dimension-based category ranking relationship table, a category mapping control relationship table, a sales script conversion processing table, a model prediction result correction table, and a marketing personnel actual result feedback correction table, so as to construct the model result interpretation system through the interrelationship between the tables.
[0169] This embodiment constructs a global category ranking relationship table, a dimension-based category ranking relationship table, a category mapping control relationship table, a sales conversion processing table, a model prediction result correction table, and a marketing personnel actual result feedback correction table. Through the interrelationships between these tables, a model result interpretation system is built. This bridges the technical gap between the complex mathematical principles behind the model and business personnel unfamiliar with it. It helps business personnel accurately identify marketing entry points and determine marketing methods in specific usage scenarios by leveraging model prediction results, their own experience, and the model result interpretation system, thereby improving marketing efficiency, accuracy, and the lower limit of marketing effectiveness. Furthermore, the model prediction results can be corrected based on actual marketing scenarios and marketing personnel feedback. Incorrectly predicted samples can be incorporated into the next model update iteration, gradually improving the accuracy of the marketing model.
[0170] Reference Figure 4 , Figure 4 This is a flowchart illustrating the third embodiment of the marketing model data processing method of the present invention.
[0171] like Figure 4 As shown, the embodiments of the present invention are based on the above. Figure 3 In the embodiment shown, before step S103 above, which involves predicting the multi-dimensional marketing category data based on a preset marketing model to obtain the model prediction result for the marketing scenario, the following steps are also included:
[0172] Step S1001: Construct the marketing model.
[0173] Compared to the above Figure 3 The embodiment shown also includes a scheme for constructing a marketing model.
[0174] By constructing a complete model training and iteration method, the goal is to continuously optimize the model and improve its accuracy.
[0175] Specifically, the marketing model can be trained and iteratively constructed using the following method:
[0176] Step 1, construct the initial dataset;
[0177] One approach is to construct an initial dataset using human experience and rules.
[0178] Step 2: Train the model using the initial dataset and validate it to obtain the trained initial model;
[0179] Step 3: Use the trained initial model to predict the actual scene data and obtain the prediction results of the initial model;
[0180] For example, using a pre-trained model to predict marketing results for the next month.
[0181] Step 4: Compare the initial model prediction results with the actual results, and determine the final initial model prediction results based on the comparison results;
[0182] Specifically, predictions can be made on the entire dataset. The predictions from the initial model are compared with the actual results, and the one with the higher result is taken as the final prediction from the initial model.
[0183] The actual result is obtained through processing according to certain rules; predicting the full dataset involves predicting the results for all existing users in the system.
[0184] The reason for comparing the initial model prediction results with the actual results is to implement a fallback strategy to prevent model anomaly detection and avoid obvious erroneous prediction results.
[0185] Step 5: Based on the final initial model prediction results, perform stratified sampling on the data of the model's prediction errors, and use it as a supplementary training dataset for the next round of model iteration, so that the model can focus more on the samples of prediction errors.
[0186] Step 6: Based on the final initial model prediction results, perform stratified sampling on the data that the model predicted correctly, and use it as the training dataset for the next round of model iteration.
[0187] Among them, stratified sampling refers to sampling different numbers of customers at different levels.
[0188] For example, 100,000 data points are sampled for Level 1 users, and 200,000 data points are sampled for Level 2 users.
[0189] Step 7: Use the training dataset and the training supplementary dataset as the data basis for the next iteration of the model, and iterate the model.
[0190] Step 8: Repeat steps 3-7 above until the expected stopping condition is met, then end the model iteration and obtain the trained marketing model.
[0191] The expected stopping condition can be a pre-set number of iterations or the value of the loss function. When the number of iterations or the value of the loss function reach the preset condition, the iteration can be terminated, and the trained marketing model can be obtained.
[0192] This embodiment, through the above-described scheme, specifically acquires marketing information by constructing a marketing model and a model result interpretation system; performs data mining on the marketing object based on the marketing information to obtain multi-dimensional category data reflecting the marketing object; predicts the multi-dimensional category data based on the preset marketing model to obtain model prediction results under the marketing scenario; and interprets the model prediction results based on the pre-constructed model result interpretation system to obtain and display prominent marketing categories under different dimensions. This invention focuses on solving the confusion faced by business personnel when using model results. By combining specific business scenarios, it constructs a complete and highly interpretable model result interpretation system, and intuitively and dynamically displays prominent marketing categories under different dimensions on the front-end page, enabling business personnel to combine model prediction results with their own experience to make targeted marketing decisions. This embodiment can bridge the technical gap between the complex mathematical principles behind the model and business personnel unfamiliar with the model, helping them accurately find marketing entry points and determine marketing methods in specific usage scenarios through multiple dimensions such as model prediction results, their own experience, and the model result interpretation system, thereby improving marketing efficiency, accuracy, and the lower limit of marketing effectiveness. Furthermore, the model prediction results can be corrected based on actual marketing scenarios and feedback from marketers. Incorrectly predicted samples can be incorporated into the next update and iteration of the model, thereby gradually improving the accuracy of the marketing model.
[0193] Furthermore, based on the above scheme, a two-dimensional relational database can be constructed to store relevant information, such as user-related information and model prediction results, which can then serve as the data foundation for subsequent training and model iteration.
[0194] Furthermore, embodiments of the present invention also propose a marketing model data processing system, the system comprising:
[0195] The acquisition module is used to acquire marketing information;
[0196] The data mining module is used to perform data mining on the marketing object based on the marketing information to obtain multi-dimensional category data reflecting the marketing object;
[0197] The prediction module is used to predict the multi-dimensional category data of marketing based on a preset marketing model, and obtain the model prediction results in the marketing scenario.
[0198] The explanation and display module is used to interpret the model prediction results based on the pre-built model result interpretation system, and to obtain and display the prominent marketing categories under different dimensions.
[0199] The principle of marketing model data processing implemented in this embodiment can be referred to in the above embodiments, and will not be repeated here.
[0200] Furthermore, this invention also proposes a marketing model data processing system, which includes: a memory, a processor, and a marketing model data processing program stored in the memory and executable on the processor. When the marketing model data processing program is executed by the processor, it implements the marketing model data processing method as described in the above embodiments.
[0201] The principle of marketing model data processing implemented in this embodiment can be referred to in the above embodiments, and will not be repeated here.
[0202] Furthermore, embodiments of the present invention also propose a computer-readable storage medium storing a marketing model data processing program, which, when executed by a processor, implements the marketing model data processing method as described in the above embodiments.
[0203] The principle of marketing model data processing implemented in this embodiment can be referred to in the above embodiments, and will not be repeated here.
[0204] Compared to existing technologies, this invention proposes a marketing model data processing method, system, and computer-readable storage medium. The method involves acquiring marketing information; performing data mining on the marketing object based on the marketing information to obtain multi-dimensional category data reflecting the marketing object; predicting the multi-dimensional category data based on a preset marketing model to obtain model prediction results for the marketing scenario; and interpreting the model prediction results based on a pre-built model result interpretation system to obtain and display prominent marketing categories under different dimensions. This invention focuses on addressing the confusion faced by business personnel when using model results. By combining specific business scenarios, it constructs a complete and highly interpretable model result interpretation system, and intuitively and dynamically displays prominent marketing categories under different dimensions on a front-end page. This allows business personnel to combine model prediction results with their own experience to make targeted marketing decisions. This embodiment bridges the technical gap between the complex mathematical principles behind the model and business personnel unfamiliar with it. It helps business personnel accurately identify marketing entry points and determine marketing methods in specific use cases by leveraging model predictions, their own experience, and the model's interpretable framework. This aims to improve marketing efficiency, accuracy, and the lower limit of marketing effectiveness. Furthermore, the model's predictions can be revised based on actual marketing scenarios and feedback from marketing personnel. Incorrect predictions can be incorporated into subsequent model updates, gradually improving the accuracy of the marketing model.
[0205] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.
[0206] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.
[0207] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a recommendation effect evaluation system (which may be a mobile phone, computer, server, controlled terminal, or network device, etc.) to execute the methods of each embodiment of the present invention.
[0208] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.
Claims
1. A marketing model data processing method, characterized in that, The method includes the following steps: Obtain marketing information; Data mining is performed on the marketing targets based on the aforementioned marketing information to obtain multi-dimensional marketing category data reflecting the marketing targets. Based on a preset marketing model, the multi-dimensional category data of marketing is predicted to obtain the model prediction results in the marketing scenario. The model prediction results are interpreted based on a pre-built model result interpretation system to obtain and display prominent marketing categories under different dimensions; Before the step of interpreting the model prediction results based on the pre-built model result interpretation system to obtain and display the prominent marketing categories under different dimensions, the following steps are also included: Acquire marketing scenario data; The marketing scenario data is anonymized. Based on the marketing scenario data, a marketing profile combining dynamic and static elements is constructed through multiple dimensions; each dimension includes several subcategories, and each subcategory includes various user characteristics. Based on the aforementioned marketing profile, a model result interpretation system is obtained to provide an intuitive interpretation of the model prediction results; The steps for constructing a marketing profile combining dynamic and static elements based on the marketing scenario data across multiple dimensions include: Based on the marketing scenario data, obtain static user characteristics from multiple dimensions, and construct static profiles based on the static user characteristics; Based on the marketing scenario data, dynamic user characteristics in multiple dimensions are obtained. Based on these dynamic user characteristics, a dynamic profile is constructed through global feature comparison and intra-dimensional feature comparison. A marketing profile is constructed by combining the static and dynamic profiles. The steps of obtaining dynamic user characteristics across multiple dimensions based on the marketing scenario data, and constructing a dynamic profile based on these dynamic user characteristics through global feature comparison and intra-dimensional feature comparison, include: Based on the marketing scenario data, dynamic user characteristics of different categories in multiple dimensions are obtained; For each dynamic user feature, perform a global vertical comparison of the data, calculate the ranking of a single data point in the global data within the current category, and obtain the percentage ranking of a single data point within the current category; The top-ranked data points are used to create a dynamic global profile. Based on the percentage of data obtained from global vertical comparison, the categories within each dimension of each data point are filtered to remove invalid, abnormal, and empty categories; After filtering each data point, the categories within each dimension are compared horizontally, and the top-performing categories by percentage are displayed to obtain a different dynamic profile for each data point. The step of obtaining a model result interpretation system based on the marketing profile to intuitively interpret the model prediction results also includes: Establish a global category ranking relationship table, a dimension-based category ranking relationship table, a category mapping control relationship table, a sales script conversion processing table, a model prediction result correction table, and a marketing personnel actual result feedback correction table, so as to construct the model result interpretation system through the interrelationship between the tables.
2. The method according to claim 1, characterized in that, The method further includes: Obtain feedback from marketing personnel; The model's predictions were revised based on feedback from the marketers.
3. The method according to claim 1, characterized in that, The subcategories are divided into numerical and character types. The step of acquiring dynamic user characteristics across multiple dimensions based on the marketing scenario data, and constructing a dynamic profile based on these dynamic user characteristics through global feature comparison and intra-dimensional feature comparison, further includes: For the numerical categories in the marketing scenario data, calculate the global ranking percentage of the marketing object; For the character-type categories in the marketing scenario data, the character-type categories are transformed, and each character-type category is assigned a corresponding weight through statistical methods. The ranking percentage of the sub-category is calculated based on the weight.
4. The method according to claim 1, characterized in that, The steps of obtaining a model result interpretation system based on the marketing profile to intuitively interpret the model prediction results include: Based on the aforementioned marketing profile, a communication script system is constructed to process the displayed categories using professional communication techniques.
5. The method according to any one of claims 1-4, characterized in that, Before the step of predicting the multi-dimensional category data of marketing based on a preset marketing model to obtain the model prediction results in the marketing scenario, the following steps are also included: The marketing model is constructed by including: Build the initial dataset; The model is trained and validated using the initial dataset to obtain a trained initial model. The trained initial model is used to predict real-world scene data to obtain the initial model prediction results; The initial model prediction results are compared with the actual results, and the final initial model prediction results are determined based on the comparison results. Based on the final initial model prediction results, the data with incorrect model predictions are stratified and sampled as a supplementary training dataset for the next round of model iteration. Based on the final initial model prediction results, the data that was correctly predicted by the model is stratified and used as the training dataset for the next round of model iteration. The training dataset and the training supplementary dataset are used as the data basis for the next iteration of the model. The model is iterated until the expected stopping condition is met, and the model iteration ends, resulting in a trained marketing model.
6. A marketing model data processing system, characterized in that, The system includes: The acquisition module is used to acquire marketing information; The data mining module is used to perform data mining on the marketing object based on the marketing information to obtain multi-dimensional marketing category data that reflects the marketing object; The prediction module is used to predict the multi-dimensional category data of marketing based on a preset marketing model, and obtain the model prediction results in the marketing scenario. The explanation and display module is used to explain the model prediction results based on the pre-built model result explanation system, and to obtain and display the prominent marketing categories under different dimensions; Before interpreting the model prediction results based on the pre-built model result interpretation system, and obtaining and displaying the prominent marketing categories under different dimensions, the process also includes: Acquire marketing scenario data; The marketing scenario data is anonymized. Based on the marketing scenario data, a marketing profile combining dynamic and static elements is constructed through multiple dimensions; each dimension includes several subcategories, and each subcategory includes various user characteristics. Based on the aforementioned marketing profile, a model result interpretation system is obtained to provide an intuitive interpretation of the model prediction results; The construction of a marketing profile combining dynamic and static elements based on the marketing scenario data across multiple dimensions includes: Based on the marketing scenario data, obtain static user characteristics from multiple dimensions, and construct static profiles based on the static user characteristics; Based on the marketing scenario data, dynamic user characteristics in multiple dimensions are obtained. Based on these dynamic user characteristics, a dynamic profile is constructed through global feature comparison and intra-dimensional feature comparison. A marketing profile is constructed by combining the static and dynamic profiles. The process of acquiring dynamic user characteristics across multiple dimensions based on the marketing scenario data, and constructing a dynamic profile based on these dynamic user characteristics through global feature comparison and intra-dimensional feature comparison, includes: Based on the marketing scenario data, dynamic user characteristics of different categories in multiple dimensions are obtained; For each dynamic user feature, perform a global vertical comparison of the data, calculate the ranking of a single data point in the global data within the current category, and obtain the percentage ranking of a single data point within the current category; The top-ranked data points are used to create a dynamic global profile. Based on the percentage of data obtained from global vertical comparison, the categories within each dimension of each data point are filtered to remove invalid, abnormal, and empty categories; After filtering each data point, the categories within each dimension are compared horizontally, and the top-performing categories by percentage are displayed to obtain a different dynamic profile for each data point. The model result interpretation system based on the marketing profile, which provides an intuitive interpretation of the model prediction results, also includes: Establish a global category ranking relationship table, a dimension-based category ranking relationship table, a category mapping control relationship table, a sales script conversion processing table, a model prediction result correction table, and a marketing personnel actual result feedback correction table, so as to construct the model result interpretation system through the interrelationship between the tables.
7. A marketing model data processing system, characterized in that, The marketing model data processing system includes: a memory, a processor, and a marketing model data processing program stored in the memory and executable on the processor. When the marketing model data processing program is executed by the processor, it implements the marketing model data processing method as described in any one of claims 1 to 5.
8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a marketing model data processing program, which, when executed by a processor, implements the marketing model data processing method as described in any one of claims 1 to 5.