Marketing method and device based on funnel analysis, storage medium and electronic equipment

Abstract

The invention provides a marketing method and device based on funnel analysis, a storage medium and electronic equipment. The method comprises the steps that a basic sample data set is acquired; extracting a plurality of keywords from the basic sample data set, inputting all the keywords into a feature extraction model to obtain a plurality of customer group topics and keywords corresponding to the customer group topics, and labeling topic tags for the customer data according to all the keywords; adopting a clustering algorithm to perform clustering analysis on all the customer data according to the theme labels of the customer data to obtain at least one customer data set and corresponding customer data; and writing all the data into the Dores database in batches, and querying the data through the Dores database to perform funnel analysis. The method combines the advantages of a feature extraction model and a clustering algorithm, orderly combines the input and output of the algorithms, and greatly improves the accuracy of customer group positioning through the cross validation of the prediction results of the two algorithms.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of marketing, in particular to a product marketing method based on funnel analysis, a product marketing device based on funnel analysis, a computer readable storage medium and an electronic device. BACKGROUND

[0002] With the rapid development of information technology, big data is increasingly widely used in the field of marketing. As an important means of evaluating the effect of marketing activities, product marketing based on funnel analysis puts higher requirements on data processing and analysis. However, the product marketing device based on funnel analysis in the traditional BI tool often faces performance bottlenecks when processing large-scale and high-concurrency data, and cannot meet the needs of enterprises for real-time and accurate analysis.

[0003] The existing marketing funnel analysis method has the problems of solidified overall marketing scheme, lack of attraction, repeated marketing to customers at the same level, and a large number of recommendations of low-level products at the same level, which cannot meet the needs of enterprise marketing for real-time data analysis and cannot accurately recommend marketing products to customers. SUMMARY

[0004] The main purpose of the present application is to provide a product marketing method based on funnel analysis, a product marketing device based on funnel analysis, a computer readable storage medium and an electronic device, so as to at least solve the problem that the data processing capacity of the existing funnel analysis tool is low, resulting in delayed analysis results and inability to accurately recommend marketing products to customers.

[0005] In order to achieve the purpose, according to one aspect of the present application, a product marketing method based on funnel analysis is provided, comprising: obtaining a basic sample data set, the basic sample data set comprising customer data of a plurality of customers, the customer data comprising historical marketing activity data and customer information of the customers; extracting a plurality of keywords from the basic sample data set, and inputting all the keywords into a feature extraction model to obtain a plurality of customer group topics and corresponding keywords, and labeling each customer data with a topic label according to the keywords of the customer group topic; performing clustering analysis on all the customer data according to the topic labels of the customer data by using a clustering algorithm to obtain a plurality of customer data sets, each customer data set comprising a plurality of customer data; writing the customer data of the customer data set and the corresponding topic labels and keywords into a Doris database, performing analysis on all the data in the Doris database by using funnel analysis to obtain an analysis result, and pushing products to corresponding customers based on the analysis result.

[0006] Optionally, a plurality of keywords are extracted from the basic sample data set, and all the keywords are input into a feature extraction model to obtain a plurality of customer group themes and corresponding keywords, and theme labels are labeled for each customer data according to the keywords of the customer group themes, including: the data in the basic sample data set is subjected to data cleaning and word segmentation processing, and keywords related to the business are screened out, the keywords at least including: customer asset information, holding product information, customer level and other dimension data, and the keywords being one of Chinese characters, numbers and English; a word segmentation dictionary is constructed, and the number of occurrences of each keyword is counted based on the word segmentation dictionary; the keywords and the corresponding number of occurrences are input into the feature extraction model to obtain model output data, the model output data at least including the customer group theme, the keywords corresponding to the customer group theme, and the weight proportion of the keywords corresponding to the customer group theme; theme labels are labeled for each customer data according to the model output data, and the number of customers of each customer group theme is determined according to the theme label, the keywords and the weight proportion of the keywords of each customer data.

[0007] Optionally, the keywords and the corresponding number of occurrences are input into the feature extraction model to obtain model output data, including: the feature extraction model is constructed based on a Spark MLlib machine learning engine, and a preset number of keywords is obtained; the keywords and the corresponding number of occurrences are input into the feature extraction model, and the customer group theme, the keywords corresponding to the customer group theme, and the weight proportion of the keywords corresponding to the customer group theme are calculated based on the preset number of keywords using an Mallet version of LDA algorithm to obtain the model output data.

[0008] Optionally, a clustering algorithm is used to perform clustering analysis on all the customer data according to the theme labels of the customer data to obtain a plurality of customer data sets, each of the customer data sets including a plurality of customer data, including: the format of the customer data and the corresponding theme labels is converted into a target format to obtain a converted data set; the converted data set is input into a K-Means clustering algorithm for iterative training to obtain an optimal number of customer data sets and customer data of each customer data set.

[0009] Optionally, all data is batch written into a Doris database, including: obtaining the Doris database, wherein an interface is configured to associate a raw table and a target table through the Doris database, the raw table is a data table including all customer data and corresponding topic labels and keywords, and the target table is a query data table; all data is batch written into the Doris database through a spark task, the customer data and the corresponding topic labels and keywords are configured into a Doris data engine to establish an index, and a unique model is created, taking a customer unique number and other key fields as a primary key, to obtain the Doris database after writing data.

[0010] Optionally, funnel analysis is used to analyze all data in the Doris database to obtain an analysis result, including: obtaining a target screening condition, the target screening condition being composed of at least one screening logic and at least one sub-screening condition; defining an index that needs to be aggregated and counted according to the target screening condition and a business requirement; and performing associated query on corresponding Doris data according to the index that needs to be aggregated and counted, and sending parsed data after query to a target page.

[0011] Optionally, a basic sample data set is obtained, including: determining a target collection mode, the target collection mode including at least one of the following: mapping to a storage table based on a message queue, directly reading and collecting through an external heterogeneous data storage RPC, collecting a text file, collecting through a real-time transaction API interface, and federated collecting a HIVE data warehouse; collecting initial sample data by using the target collection mode, the initial sample data including real-time dynamic data and basic static data, the basic static data including at least customer information, and the real-time dynamic data including at least one of the following: asset details, user behavior data, and real-time asset change information; and performing associated and summarized processing on the basic static data and the real-time dynamic data through a Spark preprocessing program to obtain a plurality of customer data, each customer data taking a customer number as a unique primary key.

[0012] According to another aspect of the present application, there is provided a funnel analysis-based product marketing apparatus, comprising: an acquisition unit configured to acquire a basic sample dataset, the basic sample dataset comprising customer data of a plurality of customers, the customer data comprising historical marketing activity data and customer information of the customers; a feature extraction unit configured to extract a plurality of keywords from the basic sample dataset, input all the keywords into a feature extraction model, obtain a plurality of customer group topics and corresponding keywords, and label each of the customer data with a topic label according to the keywords of the customer group topics; a clustering unit configured to perform clustering analysis on all the customer data according to the topic labels of the customer data using a clustering algorithm, and obtain a plurality of customer data sets, each of the customer data sets comprising a plurality of the customer data; and an analysis unit configured to write the customer data of the customer data sets and corresponding topic labels and keywords into a Doris database, perform analysis on all the data in the Doris database using a funnel analysis to obtain an analysis result, and push a product to corresponding customers based on the analysis result.

[0013] According to another aspect of the present application, there is provided a computer-readable storage medium comprising a stored program, wherein the program, when executed, controls a device in which the computer-readable storage medium is located to perform any of the funnel analysis-based product marketing methods.

[0014] According to another aspect of the present application, there is provided an electronic device comprising one or more processors, memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs comprise a program for performing any of the funnel analysis-based product marketing methods.

[0015] With the technical solution of the application, the product marketing method based on funnel analysis first acquires a basic sample data set; extracts a plurality of keywords from the basic sample data set, inputs all the keywords into a feature extraction model, obtains a plurality of customer group themes and keywords corresponding to each customer group theme, and labels each customer data with a theme label according to all the keywords; then uses a clustering algorithm to perform clustering analysis on all the customer data according to the theme label of the customer data, and obtains at least one customer data set and corresponding customer data; finally, all the data are written into a Doris database in batches, and the data are queried through the Doris database to perform funnel analysis. The method combines the advantages of the feature extraction model and the clustering algorithm, orderly combines the input and output of the algorithm, and cross- validates the predicted results of the two algorithms, greatly improving the accuracy of customer group positioning, solving the problem of low data processing capacity of the funnel analysis tool in the prior art, causing delay of the analysis result, and being unable to recommend accurate marketing products to customers. BRIEF DESCRIPTION OF DRAWINGS

[0016] The drawings accompanying the specification of the present application form a part thereof, serve to provide further understanding of the present application, and together with the specification explain the application. The use of these drawings in explaining the application does not imply that the present application should be limited to the embodiments illustrated therein. In the drawings:

[0017] Figure 1 A hardware structure block diagram of a mobile terminal for performing a product marketing method based on funnel analysis is shown according to an embodiment of the present application;

[0018] Figure 2 A flowchart of a product marketing method based on funnel analysis is shown according to an embodiment of the present application;

[0019] Figure 3 A flowchart of another product marketing method based on funnel analysis is shown according to an embodiment of the present application;

[0020] Figure 4 A structure block diagram of a product marketing device based on funnel analysis is shown according to an embodiment of the present application.

[0021] Among them, the above-mentioned drawings include the following reference signs:

[0022] 102, processor; 104, memory; 106, transmission device; 108, input and output device. DETAILED DESCRIPTION

[0023] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the drawings and in combination with the embodiments.

[0024] In order to make the person skilled in the art better understand the scheme of the present application, the technical scheme in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by the person skilled in the art without creative labor should belong to the protection scope of the present application.

[0025] It should be noted that the terms "first", "second" and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0026] For the convenience of description, part of the nouns or terms related to the embodiments of the present application are described as follows:

[0027] Doris: a high-performance, real-time analytical database based on MPP architecture, which can return query results under massive data within sub-second response time, not only can support high-concurrency point query scenarios, but also can support high-throughput complex analysis scenarios, support PB-level real-time query and analysis of data, and highly compatible with mysql protocol and mysql syntax;

[0028] Funnel analysis: a method for analyzing the conversion process of user behavior, which can help understand the conversion rate of users from one stage to another in a specific process. Funnel represents the flow of users in a predefined path, and is usually used to describe the conversion rate of users in the process of completing a certain goal or achieving a certain conversion event;

[0029] Ad hoc analysis: a specific analysis aimed at solving specific problems, which can quickly find the positioning of instant business scenarios by analyzing data, and can provide flexible data analysis capabilities, such as taking data and analyzing data at any time; multi-dimensional data combination allows different dimensions to be assembled and combined to form facets, enabling more dimensional analysis, especially suitable for marketing scenarios based on big data;

[0030] Spark MLlib: Spark's scalable machine learning library, which consists of a series of machine learning algorithms and utilities. It includes classification, regression, clustering, collaborative filtering, and other algorithms, as well as some underlying optimization methods.

[0031] Feature extraction: Feature extraction algorithms are important techniques in the fields of machine learning and data mining, used to extract meaningful and representative features from raw data for subsequent model training and data analysis.

[0032] Clustering algorithm: Clustering algorithm is a commonly used unsupervised learning algorithm, which is used to divide the objects in the data set into several groups, so that the similarity within the same group is high, and the similarity between different groups is low.

[0033] Customer circle selection: Refers to selecting customer groups with similar interests or behaviors based on specific labels, attributes, or other characteristics of marketing or observation objects, in order to make more targeted and precise marketing to these customers. Refers to the selection of customer groups based on specific conditions, and statistical analysis of customers from different customer attribute dimensions according to actual business scenarios;

[0034] HIVE library: HIVE is a data warehouse tool based on a distributed file system, used for data extraction, transformation, and loading.

[0035] As introduced in the background art, the deficiencies of the prior art mainly include the following points:

[0036] 1) In the existing technology, customer data is stored in an offline data warehouse such as HIVE during big data analysis, and the distribution of customer indicators is calculated and aggregated through non-real-time methods. This approach is computationally intensive, time-consuming, and storage redundant, and cannot meet the second-level response performance requirements of real-time queries, and is difficult to cope with complex logical combination filtering of multiple conditions;

[0037] 2) Existing technologies such as ES and clickhouse only perform simple hierarchical classification on customers according to a single dimension such as assets and held products, and cannot classify, locate, and select valuable customer groups in multiple coordinate vector dimensions, resulting in relatively generalized and single marketing promotion methods that cannot effectively tap the potential value of customers;

[0038] 3) The existing technology cannot display customer basic information and asset details in all dimensions in real time, and lacks a three-dimensional holographic display of classified customer information, making it impossible to compare and analyze real-time and historical data;

[0039] 4) The existing technology limits the development of customer groups layer by layer, and the overall marketing plan is fixed and lacks appeal, resulting in repeated recommendations of low-level products to customers at the same level.

[0040] 5) Existing funnel analysis tools often face performance bottlenecks when handling large-scale, high-concurrency data. Current technologies have limited data processing capabilities; when the data volume reaches a certain level, traditional analysis tools often cannot quickly complete data loading, calculation, and visualization, leading to delays in analysis results. In a big data environment, this delay may become particularly pronounced, failing to meet the real-time data analysis needs of enterprise marketing.

[0041] Existing funnel-based product marketing tools lack scalability. As business grows and data increases, they cannot provide flexible expansion solutions, leading to cumbersome system upgrades or replacements when facing data growth, increasing additional costs and time investment.

[0042] To address the problem that existing funnel analysis tools have low data processing capabilities, resulting in delayed analysis results and an inability to accurately recommend marketing products to customers, embodiments of this application provide a product marketing method based on funnel analysis, a product marketing device based on funnel analysis, a computer-readable storage medium, and an electronic device.

[0043] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0044] The methods and embodiments provided in this application can be executed on a mobile terminal, a computer terminal, or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 This is a hardware structure block diagram of a mobile terminal for a product marketing method based on funnel analysis, according to an embodiment of the present invention. Figure 1 As shown, a mobile terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 (which may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.) and a memory 104 for storing data are also shown. The mobile terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the mobile terminal described above. For example, the mobile terminal may also include components that are more... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0045] The memory 104 can be used to store computer programs, such as software programs of application software and modules, such as a computer program corresponding to the funnel analysis-based product marketing method in the embodiments of the present application. The processor 102 performs various functional applications and data processing by running the computer programs stored in the memory 104, that is, implements the above method. The memory 104 can include a high-speed random access memory, and can further include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some examples, the memory 104 can further include memories remotely arranged with respect to the processor 102, which can be connected to the mobile terminal through a network. Examples of the above network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof. The transmission device 106 is used to receive or send data via a network. The specific examples of the above network can include a wireless network provided by a communication provider of the mobile terminal. In one example, the transmission device 106 includes a network adapter (NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet in a wireless manner.

[0046] In the embodiments, a funnel analysis-based product marketing method running on a mobile terminal, a computer terminal or a similar computing device is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that herein.

[0047] Figure 2 is a flowchart of the funnel analysis-based product marketing method according to the embodiments of the present application. As shown in Figure 2 , the method includes the following steps:

[0048] In step S201, a basic sample data set is obtained, the basic sample data set including customer data of a plurality of customers, and the customer data including historical marketing activity data and customer information of the customers;

[0049] Specifically, the traditional marketing method has high commonality, lacks real-time, intelligent and multi-dimensional analysis and integration of the relationship between inventory customers and existing products, and the marketing scheme is relatively single. Most of them use the existing technical framework, based on the simple classification of products, carry out feature extraction from a single dimension, the data is not time-effective, which leads to a large number of customer loss, and the precise promotion effect of marketing resources is not obvious.

[0050] The above steps are driven by customer insight, based on funnel analysis mode, and real-time multi-dimensional data analysis of customers, building specific customer groups, using existing data to analyze customer behavior distribution, multi-dimensional and multi-condition filtering of specific precise customer groups, and using customized marketing solutions to carry out "precise drip irrigation".

[0051] Generally, customer data includes customer information, asset details, tag information, user behavior data, real-time asset change data, and the like.

[0052] The method comprises the following steps:

[0053] Step S2011, determining a target collection method, wherein the target collection method comprises at least one of the following: mapping to a storage table based on a message queue, external heterogeneous data storage RPC direct reading collection, text file collection, real-time transaction API interface collection, and HIVE data warehouse federation collection.

[0054] Step S2012, collecting initial sample data by using the target collection method, wherein the initial sample data comprises real-time dynamic data and basic static data, the basic static data at least includes customer information, and the real-time dynamic data at least includes one of the following: asset details, user behavior data, and real-time asset change data.

[0055] Step S2013, associating and summarizing the basic static data and the real-time dynamic data by using a Spark preprocessing program to obtain a plurality of customer data, and each customer data takes a customer number as a unique primary key.

[0056] Specifically, the conventional marketing means performs simple screening and filtering based on existing offline data in a relational database, and then combines the existing marketing scheme to perform single-means marketing on customers. First, the analysis and mining of the basic data are low, and most of them are simple basic information table association. The customer granularity of screening is coarse, the range is not accurate, and it is more difficult to organically combine the basic information, asset information, tags, and the like of the customer holographic dimension.

[0057] The above steps collect mobile phone data from various marketing channels and perform preliminary data cleaning, integration and preprocessing to provide basic data for subsequent data analysis. In data collection, the above steps provide multiple collection methods based on message queue mapping to storage table, external heterogeneous data storage RPC direct reading collection, text file collection, real-time transaction API interface collection, HIVE data warehouse federation collection, etc. The collected data includes customer information, asset details, label information, user behavior data, real-time asset change data, etc. The massive static basic data and real-time data are associated and summarized by the Spark preprocessing program, each data is taken as a unique primary key by the customer number, and the required data set is obtained.

[0058] Step S202, extracting a plurality of keywords from the above basic sample data set, and inputting all the above keywords into a feature extraction model to obtain a plurality of customer group themes and corresponding keywords, and labeling each of the above customer data with a theme label according to the keywords of the above customer group theme;

[0059] Specifically, feature extraction has been widely applied in various data analysis and mining, and has a wide range of applications in marketing scenarios. It can help better understand customer groups, optimize marketing strategies and improve market effectiveness, such as customer segmentation, market segmentation, product recommendation, customer churn and other marketing methods. Feature extraction algorithms can extract key features from customer data, such as purchase frequency, unsubscribe status, etc. By using feature extraction algorithms to mine the behavior trends of the most valuable customer groups, the operation of the customer group is analyzed, which facilitates marketing personnel to make real-time and accurate marketing strategies based on the prediction list, so as to maintain the customer data set and improve the customer level layer by layer.

[0060] Among them, a plurality of keywords are extracted from the above basic sample data set, and all the above keywords are input into a feature extraction model to obtain a plurality of customer group themes and corresponding keywords, and each of the above customer data is labeled with a theme label according to the keywords of the above customer group theme, including the following steps:

[0061] Step S301, data cleaning and word segmentation processing are performed on the data in the above basic sample data set, and keywords related to the business are selected, the above keywords at least including: customer asset information, product information, customer level data in the dimension, and the above keywords being one of Chinese characters, numbers and English;

[0062] Step S302, constructing a word segmentation dictionary and counting the number of occurrences of each of the above keywords based on the above word segmentation dictionary;

[0063] Step S303, input the keywords and corresponding occurrence times into the feature extraction model to obtain model output data, wherein the model output data at least includes the customer group theme, keywords corresponding to the customer group theme, and weight proportion of the keywords corresponding to the customer group theme;

[0064] The feature extraction model is constructed based on a Spark MLlib machine learning engine, and a preset number of keywords is obtained.

[0065] Step S3031, the feature extraction model is constructed based on a Spark MLlib machine learning engine, and a preset number of keywords is obtained.

[0066] Step S3032, input the keywords and corresponding occurrence times into the feature extraction model, and calculate the customer group theme, keywords corresponding to the customer group theme, and weight proportion of the keywords corresponding to the customer group theme based on the preset number of keywords using the Mallet version of the LDA algorithm to obtain the model output data, wherein the data format of the model output data is [theme ID, theme Distribution (keyword weight x keyword)].

[0067] Specifically, the customer behavior features of the massive original customer sample data in the HIVE library are extracted by the feature extraction model using the Spark MLlib platform.

[0068] The basic sample data set is cleaned and segmented, and Chinese characters, numbers and English related to the business are selected, mainly including customer asset information, product information, label information, customer level and other dimension data keywords; then a segmentation dictionary is constructed, the number of occurrences of each keyword is counted as the model input; finally, a feature model is constructed, according to the previous business experience accumulation, the number of key themes is set to 10, based on the Spark MLlib machine learning engine, after multiple attempts, it is determined to use the Mallet version of the LDA algorithm to improve the original LDA model, and finally the preliminary key theme and its detailed distribution are calculated, 10 keywords that best represent each theme and the weight proportion of each keyword are extracted, and the data format is [theme ID, theme Distribution (keyword weight x keyword)].

[0069] Step S304, according to the model output data, each customer data is labeled with a theme label, and the number of customers of each customer group theme is determined according to the theme label, keywords, and weight proportion of the keywords of each customer data.

[0070] Based on the feature model output result, it is verified and confirmed that 10 themes are less than the optimal theme number by using the one-time score mechanism, and the initially selected theme number does not need to be adjusted; with the determined theme, each customer data is labeled with the theme, and the keywords and the score proportion of the data under the theme label are calculated; after each data is labeled with the theme label by the feature extraction model, the number of customers and the proportion of each theme are counted and summarized, and the customer situation of each theme can be directly viewed.

[0071] Specifically, the feature extraction model first performs data cleaning, word segmentation, and screening of keywords related to the business on the massive customer information, constructs a word segmentation dictionary and counts the number of occurrences of keywords. Then, based on the accumulated business experience in the early stage, the number of key themes is set and the LDA algorithm is used to extract the theme and the keyword weight proportion. The number of themes is verified by the one-time score mechanism, each data is labeled with a theme label, and the number of customers and the distribution of each theme are counted.

[0072] Step S203, using a clustering algorithm to perform clustering analysis on all the customer data according to the theme label of the customer data, to obtain a plurality of customer data sets, each of the customer data sets including a plurality of customer data;

[0073] Specifically, feature extraction and clustering algorithms are currently widely used in various data analysis and mining, and have a wide range of applications in marketing scenarios, which can help better understand customer groups, optimize marketing strategies, and improve market effectiveness, such as customer segmentation, market segmentation, product recommendation, and customer churn. The feature extraction algorithm can extract key features from customer data, such as purchase frequency and subscription status, and in combination with the clustering algorithm, it can identify potential customer churn groups and take targeted measures, such as launching promotional activities or personalized services, to reduce customer churn rate. The current mainstream feature extraction algorithms include LDA topic model, TD-IDF, and clustering algorithms such as K-Means algorithm, GMM, and distributed clustering algorithm.

[0074] Among them, the clustering algorithm is used to perform clustering analysis on all the customer data according to the theme label of the customer data, to obtain a plurality of customer data sets, each of the customer data sets including a plurality of customer data, including the following steps:

[0075] Step S2031, converting the format of the customer data and the corresponding theme label into a target format to obtain a converted data set, and the target format is: [keyword, keyword feature];

[0076] Step S2032: The transformed dataset is used as input to the K-Means clustering algorithm for iterative training to obtain the optimal number of customer data sets and the customer data corresponding to each customer data set.

[0077] Specifically, given the massive size of the basic user data sample, and the fact that traditional clustering algorithms are very time-consuming, the aforementioned clustering algorithm uses the Mini Batch K-Means algorithm. This algorithm uses a subset of samples from the sample set to perform traditional K-Means, thus avoiding the computational difficulties when the sample size is too large and significantly accelerating the convergence speed of the algorithm.

[0078] Since the transformed dataset needs to be used as input for the K-Means clustering algorithm, the dataset schema output by the feature extraction model needs to be converted into the form of [label, features]. That is, the topic distribution column name needs to be converted into features. In other words, the above customer data and the corresponding topic labels need to be converted into the target format, which is: [keyword, keyword feature].

[0079] In some embodiments, the maximum number of iterations is set to 200, and the number of clusters is set to 2, 4, 6, 8, 10, 12, 14, 16, 18, or 20; the default values ​​are used for the rest. The Euclidean distance between the data object and the cluster center Ci is calculated. For i = 1, 2, ..., m, the sample xi and each centroid vector μ are calculated. j Distance d (j = 1, 2, ..., k) ij =||x i -μ j ||22, x i The smallest one is d. ij The corresponding category λ i At this point, update Cλ. i =Cλ i ∪{x i}; For j = 1, 2, ..., k, for C j Recalculate the new centroid μ for all sample points. j =1|C j |∑x∈C j x; If none of the k centroid vectors change, repeat the above steps and finally observe the changes in the evaluation index.

[0080] After training, it was found that the clustering effect was best when K=4. Using the above result set and combined with the accumulation of previous business experience, and by associating basic customer data, four customer data sets were finally identified: corporate customer group, quality elderly care customer group, high-end credit card customer group, and rural private financial customer group in counties.

[0081] Step S204, write the customer data of the customer data set and the corresponding subject label and keyword into the Doris database, analyze all data in the Doris database by funnel analysis to obtain an analysis result, and push a product to the corresponding customer based on the analysis result.

[0082] Specifically, the conventional marketing means performs simple screening and filtering based on existing offline data in a relational database, and then performs single-means marketing on customers in combination with an existing marketing scheme. First, the analysis and mining of basic data are at a low level, mostly simple basic information table association, and the customer granularity of screening is coarse and the range is not accurate. It is more difficult to organically combine the basic information, asset information, labels, etc. of the customer holographic dimension. Second, the customers to be marketed cannot form a key customer group, and marketing is mostly carried out by using the existing marketing scheme, which does not meet the asset promotion needs of specific customer data sets, and the marketing means are similar, and the customers are not interested. Third, the attention to the entire asset management situation of the customers is not enough, and a long-term, stable and effective tracking and maintenance has not been formed, resulting in low customer satisfaction. Finally, there is a lack of effective means to promote the customer level layer by layer, it is difficult to realize the marketing value of the enterprise, and most of the customer levels remain unchanged for a long time, and the asset and service satisfaction are not significantly improved. If this continues, the customers will be tired of the products and services, which may cause customer loss.

[0083] The conventional solution requires the user to have a certain programming or database query language basis, converts the customer screening conditions required by the business into an executable program, and submits a job to the big data cluster. The disadvantages are obvious. On the one hand, data storage is repeated, basic data needs to be stored in multiple tables and multiple dimensions, and multiple development queries, analysis and other services are required, the entire development cycle is long and difficult, and the maintenance cost is high. On the other hand, the business logic of multi-dimension and multi-condition combination needs to be submitted to the big data cluster for calculation, and in the case of massive data, it takes several minutes or even several hours to obtain the calculation result, and the business analysis efficiency is low.

[0084] Doris provides fast query capabilities and real-time computing functions, which can meet the needs of real-time OLAP analysis, support fast data processing and query. Horizontal expansion and high reliability: Doris supports horizontal expansion, which can easily expand the cluster size to cope with data growth, while having high reliability and fault tolerance. Multi-dimensional analysis: Doris supports multi-dimensional data analysis, which can perform complex multi-dimensional queries and analysis operations to meet the needs of different business scenarios. Low-latency query: Doris provides low-latency query capabilities, which can complete large-scale data queries and analysis within seconds, suitable for real-time monitoring and decision support scenarios. Flexible data model: Doris supports flexible data models that can adapt to different data structures and formats, while providing rich data processing and conversion functions. In summary, Doris has strong advantages in real-time computing, data analysis and scalability, and is suitable for scenarios that require real-time processing and analysis of large-scale data.

[0085] wherein all data is batch written into the Doris database, including the following steps:

[0086] Step S401, obtaining the above-mentioned Doris database, wherein an interface is configured to realize the association between the original table and the target table through the above-mentioned Doris database, the original table is a data table including all the above-mentioned customer data and corresponding above-mentioned theme tags and keywords, and the target table is a query data table;

[0087] Step S402, batch writing all data into the above-mentioned Doris database through a spark task, and configuring the above-mentioned customer data and corresponding above-mentioned theme tags and keywords into the Doris data engine to establish an index, and creating a unique model with the customer unique number and other key fields as the primary key, to obtain the Doris database after writing data.

[0088] Specifically, by building a self-developed Doris batch data synchronization framework, with the help of a visual task scheduling tool, data is batch written into the Doris database through a spark task. The high concurrency, low latency real-time query and computing capabilities of Doris, as well as its powerful distributed capabilities supporting complex business analysis, are utilized to load customer detail data and theme feature data into the Doris database, establish an index, and create a unique model to improve query efficiency. The flexible data model and low-latency query function of Doris are used to realize customer group selection, precise positioning and multi-dimensional insight analysis. The horizontal expansion and high reliability of Doris enable it to cope with data growth and high-concurrency query requirements, providing a stable foundation for ad hoc queries.

[0089] In some embodiments, a self-developed Doris batch data synchronization framework is built, wherein the self-developed Doris batch data synchronization framework can associate the original table and the target table through the configuration of an interface. The original table is an original data table storing data, and the target table is a data table required to be retrieved according to the business, for example, the target table is a data table of enterprise main and guest groups in a certain region. After building the self-developed Doris batch data synchronization framework, the data is written into the Doris database in batches through the spark task with the help of the visual task scheduling tool; the customer detail dimension data and the theme feature data are loaded into the Doris data engine and indexed, and a unique model is created, taking the customer unique number and other key fields as the primary key. According to the data size, the table is divided into buckets to improve the correlation query efficiency.

[0090] Among them, the funnel analysis is used to analyze all the data in the above-mentioned Doris database to obtain an analysis result, including the following steps:

[0091] Step S501, obtaining a target screening condition, the target screening condition is composed of at least one screening logic and at least one sub-screening condition;

[0092] Step S502, defining an index that needs to be aggregated and counted according to the target screening condition and the business requirement;

[0093] Step S503, performing a correlation query on the corresponding Doris data according to the index that needs to be aggregated and counted, and sending the parsed data to the target page.

[0094] Specifically, in the screening and statistical analysis scene of a large number of customers, the excellent performance and powerful distributed capability of the Doris OLAP data engine are its greatest advantages. Compared with the traditional method, not only the response performance is greatly improved, but also the complex business analysis logic can be realized using simple and easy-to-understand SQL (database structured query prelude), and a complete set of funnel analysis scheme from customer screening to customer multi-dimensional insight is formed. The customer group selection scheme using funnel analysis effectively avoids the shortcomings of traditional marketing schemes, such as low accuracy in screening customer group range, single marketing means, and not timely follow-up in the marketing process.

[0095] In some examples, the following steps are specifically included:

[0096] 1) Multi-dimensional analysis is carried out on the determined customer data set, including all customer characteristics that need to be queried and counted, such as customer basic information and behavior labels, and customized marketing activities are carried out;

[0097] 2) Use multiple screening conditions to form a screening logic containing three logical domains of AND, OR, and NOT logical units. The AND, OR, and NOT in the unit can be composed of 0 or more specific screening conditions. The unit can also be added as a screening condition to the AND, OR, and NOT conditions in the upper logical domain. Through this downward logical combination, the business screening logic of any condition is realized;

[0098] 3) According to the customer insight dimensions of business needs, define the indicators that need to be aggregated and counted, and perform a correlation query on the corresponding Doris. Doris receives the request and calculates the indicators of each dimension required for counting. After receiving the Doris query processing result, the microservice parses and sends it to the page display;

[0099] 4) Further analyze the customer group information screened in step 2) by querying the all-around information of the customer ID: AUM and other asset conditions, tag information, product holding information, loan information, and other customer behavior information. Through the analysis of relevant indicators, the above customer group selection results are evaluated and tracked and maintained to provide high-value customers for the next step of precision marketing.

[0100] In other embodiments, a graphical customer screening condition page is provided, and the following steps are sequentially executed to assemble a complete business logic:

[0101] 1. Select a customer attribute screening condition. Customer attributes include the following three types: 1) fixed conditions, including various tags and product holding information, such as the "inactive user" tag, holding financial products, etc.; 2) fixed attribute value conditions, such as "gender = male"; 3) range attribute tags, such as "age range [20-45] years old", "asset range [10-500] million", etc.

[0102] 2. Select the logical attribute domain of the screening condition: 1) "AND", indicating that the screened customers meet all the conditions under this domain; 2) "OR", indicating that the screened customers need to meet any one of the conditions under this domain; 3) "NOT" indicating that the screened customers cannot meet any condition under this domain. Finally, combine the screening combination logic, for example:

hold 5 or more financial products AND male AND customer level is gold level and above AND (have small business owner label OR hold loan label) NOT do not disturb customers

[0103] The product marketing method based on funnel analysis provided in the present application first acquires a basic sample data set; extracts a plurality of keywords from the basic sample data set, and inputs all the keywords into a feature extraction model to obtain a plurality of customer group themes and keywords corresponding to each customer group theme, and labels each customer data with a theme tag according to all the keywords; then uses a clustering algorithm to perform clustering analysis on all the customer data according to the theme tags of the customer data to obtain at least one customer data set and corresponding customer data; finally, all the data are written into a Doris database in batches, and the data are queried through the Doris database to perform funnel analysis. The method combines the advantages of the feature extraction model and the clustering algorithm, orderly combines the input and output of the algorithm, and cross- validates the predicted results of the two algorithms, greatly improving the accuracy of customer group positioning, and solving the problem that the data processing capacity of the funnel analysis tool in the prior art is low, resulting in delayed analysis results and the inability to accurately recommend marketing products to customers.

[0104] The above embodiment uses the Spark MLlib platform to extract customer behavior features of massive original customer sample data in the HIVE library through a feature extraction model and a clustering algorithm, and the result set is output to Doris, and the target data of the basic customer portrait are used to obtain key customer group data as one of the input samples through a customer classification screening model, to mine the behavior trend of the most valuable customer group through an ad hoc analysis engine, analyze the operation of the customer group, and market to customers, for example, recommend a combination of products, so that marketing personnel can real-time and accurately market strategies based on the predicted list, thereby achieving maintenance of the customer data set and promotion of the customer level.

[0105] The above embodiment mainly performs intelligent funnel analysis marketing from the steps of customer portrait, customer positioning, operation overview, dynamic management, and promotion tracking, provides dimension logic condition nested combination, flexibly screens key customer groups, and then performs targeted marketing or analyzes the behavior distribution of customers from multiple dimensions, so that the user has a more stereoscopic customer portrait of the selected customer group, and the statistical result is output and displayed within a few seconds. For the positioned customer data set, the marketing scheme with the highest degree of fit is adopted to carry out marketing activities, and the customer level and satisfaction are accurately and effectively improved.

[0106] The above embodiment solves the problem of insufficient data growth scalability, and the present application has the characteristics of real-time dynamic elasticity, can flexibly increase and reduce data engine analysis nodes according to business needs, achieves the dual purposes of improving the processing capacity and stability of the system and reducing cost and increasing efficiency, and enables the data analysis to cope with the possible traffic peak in the marketing activities, and ensures the smooth progress of the data analysis.

[0107] The above embodiments provide displaying the analysis results to the user in the form of charts, reports, etc. Through the intuitive visual interface, the user can clearly understand the conversion of the marketing funnel, including the number of users at each stage, conversion rate, etc. indicators, providing strong support for the formulation and adjustment of marketing strategies.

[0108] In order to enable those skilled in the art to more clearly understand the technical solutions of the present application, the implementation process of the product marketing method based on funnel analysis of the present application will be described in detail below in conjunction with specific embodiments.

[0109] The present embodiment relates to a specific product marketing method based on funnel analysis, as shown in Figure 3 The first step is to collect real-time data from various marketing channels, then store the collected user basic data in the Hive library, then based on the SparkMLlib machine learning engine, perform data cleaning and word segmentation on the collected user basic data, filter out Chinese characters, numbers and English related to the business, and build a feature model. The original LDA model is improved using the Mallet version of the LDA algorithm, and the user data is classified by topic. Finally, the preliminary key topics and their detailed distribution are calculated. The output results of the feature model are input into the clustering algorithm, and the calculated customer-topic distribution is used as the input of the K-Means clustering algorithm for sample training, and finally four clustering topics are obtained. Through cross-validation to predict the final topic set, a self-developed Doris batch data synchronization framework (Doris MPP data set) is built, and through customer group selection, historical comparison, combined query and ad hoc analysis, customer portrait, customer positioning, business overview, dynamic management and upgrading tracking are used to realize intelligent funnel analysis marketing.

[0110] The key innovation of the above funnel analysis scheme lies in how to use customer portrait, customer positioning, business overview, dynamic management and upgrading tracking to realize intelligent funnel analysis marketing. By defining the logical attribute domain and combined conditions of the filtering conditions, multi-dimensional customer filtering and statistics are realized. Using the graphical customer filtering condition page, the business logic is assembled to realize accurate customer group selection and marketing activities. Through ad hoc query funnel analysis, customer behavior trend analysis, business evaluation and accurate marketing strategy formulation are realized. The overall scheme provides flexibility for dimension logic condition nesting and combination, enabling users to quickly and accurately filter out key customer groups, thereby improving marketing effectiveness and customer satisfaction.

[0111] In order to enable those skilled in the art to more clearly understand the technical solutions of the present application, the implementation process of the product marketing method based on funnel analysis of the present application will be described in detail below in conjunction with specific embodiments, including the following steps in some embodiments:

[0112] Step 1, obtain basic customer sample data.

[0113] Mobile phone data from various marketing channels are collected, and preliminary data cleaning, integration and preprocessing are performed to provide basic data for subsequent data analysis. In data collection, the present application provides multiple collection methods based on message queue mapping to storage table, external heterogeneous data storage RPC direct reading collection, text file collection, real-time transaction API interface collection, HIVE data warehouse federation collection, etc. The collected data include customer information, asset details, label information, user behavior data, real-time asset change data, etc. The massive static basic data and real-time data are associated and summarized by a Spark preprocessing program, each data is taken as a unique primary key by a customer number, and the required data set of the present patent is obtained.

[0114] Step 2, mining customer data set topic words based on feature extraction model.

[0115] 1) The result data set in step 1 is subjected to data cleaning and word segmentation, and Chinese characters, numbers and English related to the business are screened out, mainly including customer asset information, product information, label information, customer level and other dimension data keywords;

[0116] 2) A word segmentation dictionary is constructed, and the number of occurrences of each keyword is counted as the input of the model;

[0117] 3) A feature model is constructed, and according to the accumulation of previous business experience, the number of key topics is set to 10. Based on the SparkMLlib machine learning engine, after multiple attempts, the LDA algorithm of Mallet version is used to improve the original LDA model, and finally the preliminary key topics and their detailed distribution are calculated. Each topic extracts 10 keywords that best represent the topic and the weight proportion of each keyword, and the data format is [topic ID, topic Distribution (keyword weight x keyword)].

[0118] 4) Based on the output results of the above feature model, it is verified and confirmed that 10 topics are less than the best number of topics by using one-time scoring mechanism, and the initially selected number of topics does not need to be adjusted;

[0119] 5) After determining the topics, each customer data is labeled with a topic label, and the keywords under the topic and the score proportion of the data under the topic label are calculated;

[0120] 6) After each data is labeled with a topic label by the feature extraction model, the number of customers of each topic and the proportion are counted and summarized, and the customer situation of each topic can be directly observed.

[0121] Step 3: Use clustering algorithms to cluster customer groups and confirm the final theme.

[0122] Given the massive size of the basic user data sample, and after testing and verification, the traditional clustering algorithm is very time-consuming. Therefore, this patent adopts the Mini Batch K-Means algorithm, which uses a portion of the samples in the sample set to perform traditional K-Means. This avoids the computational difficulties when the sample size is too large, and the algorithm convergence speed is greatly accelerated.

[0123] 1) Since the customer-topic distribution calculated in step 2 is used as the input of the K-Means clustering algorithm, the dataset schema output by the feature extraction model needs to be converted into the form of [label, features], that is, the topic distribution column name is converted into features;

[0124] 2) Set the maximum number of iterations to 200, and the number of clusters to 2, 4, 6, 8, 10, 12, 14, 16, 18, or 20; leave the rest as default values. Calculate the Euclidean distance between the data object and the cluster center Ci. For i = 1, 2, ..., m, calculate the sample xi and each centroid vector μ. j Distance d (j = 1, 2, ..., k) ij =||x i -μ j ||22, x i The smallest one is d. ij The corresponding category λ i At this point, update Cλ. i =Cλ i ∪{x i}; For j = 1, 2, ..., k, for C j Recalculate the new centroid μ for all sample points. j =1|C j |∑x∈C j x; If none of the k centroid vectors change, repeat the above steps and finally observe the changes in the evaluation index.

[0125] 3) After training, it was found that the clustering effect was best when K=4;

[0126] 4) Using the above result set and combined with the accumulated business experience, the basic customer data was linked to finally confirm the formation of four customer data sets: corporate customer group, quality elderly care customer group, high-end credit card customer group, and rural private financial customer group in counties.

[0127] Step 4: Data synchronization with Doris.

[0128] 1) Build self-researched Doris batch data synchronization framework, use visual task scheduling tool, write data in batches to Doris database through spark task;

[0129] 2) Put customer detail dimension data and theme feature data into Doris data engine and establish index, create unique model, use customer unique number and other key fields as primary key, according to data size, divide table into buckets, and improve correlation query efficiency.

[0130] Step 5, filter customer groups and carry out funnel analysis through Doris ad hoc query.

[0131] 1) For the determined customer data set, carry out multi-dimensional analysis, including all customer characteristics that need to be queried and counted, such as customer basic information, behavior label, etc., and carry out customized marketing activities;

[0132] 2) Use multiple filtering conditions to form a filtering logic containing three logical domains of and, or, and not. The and, or, and not in the unit can be composed of 0 or more specific filtering conditions. This unit can also be added as a filtering condition to the and, or, and not conditions in the upper logical domain. Through this downward logical combination, the business filtering logic of any condition is realized;

[0133] 3) According to the customer insight dimensions of business needs, define the indicators that need to be aggregated and counted, and perform correlation query on the corresponding Doris. Doris will calculate the indicators of each dimension required for statistics after receiving the request. After receiving the Doris query processing result, the micro service parses and sends it to the page display;

[0134] 4) Further analyze the customer group information filtered in step 2) by querying the full range of information of the customer ID: AUM and other asset information, label information, product holding information, loan information, and other customer behavior information. Through the analysis of relevant indicators, the above customer group selection results are evaluated and tracked and maintained, providing high-value customers for the next step of precision marketing.

[0135] Among them, the feature extraction model and the clustering algorithm are the most widely used data analysis algorithms in the industry. In determining the customer group keywords and customer group positioning, artificial research and judgment are pre-set, and the advantages of artificial research and judgment and automatic algorithm are considered. First, the feature extraction model and the clustering algorithm can quickly process large-scale data, perform efficient feature extraction and group division, and save a lot of human resources and time cost. Second, based on data and mathematical models for analysis, the results are more objective and not affected by subjective factors, avoiding subjective errors in artificial research and judgment. Next, the feature extraction model can extract potential features from a large amount of data, and the clustering algorithm can comprehensively divide the data set, making the analysis results more comprehensive and global. Finally, these models and algorithms can adaptively adjust according to the different characteristics of the data, better adapt to various data types and scenarios, and have strong universality.

[0136] Combined with the visual task scheduling system, a self-developed Doris batch data synchronization framework is built. In the similar calculation and search or simple calculation of big data, since it is a distributed, high-extensible, and high-real-time search and data analysis database, the distributed parallel computing of Doris has absolute advantages, and the technology heat and ecology are in the rising period, which is highly suitable for the use scenario of the patent. ElasticSearch also has the ability of ad hoc query analysis, but the data comparison and code development process are relatively more complex. Under the same resource conditions, Doris is superior to ElasticSearch, so the Doris technology used in this patent can maintain a leading position for a relatively long period of time in the future, and there is no better alternative solution.

[0137] Funnel analysis is a method for analyzing user behavior flow, which is often applied to understand and optimize the user conversion process of online products or services. Compared with other analysis methods, first of all, funnel analysis can intuitively show the conversion of users at different stages, clearly display the process of user loss and conversion in the form of charts, and help analysts quickly understand the user behavior path. Secondly, through funnel analysis, key nodes and problem points of user loss can be clearly identified, helping product teams find areas that need to be improved and making targeted optimization. In addition, funnel analysis can accurately calculate the conversion rate of each conversion stage, helping enterprises quantify user conversion, and better evaluate the performance and effect of products or services. Finally, funnel analysis can help find the reasons for user loss, so as to improve product design, user experience, etc. in a targeted manner, and improve the overall user conversion rate. In general, the results of funnel analysis are based on real user data, which can help enterprises make data-driven decisions, reduce subjective judgments and assumptions, and improve the accuracy and effectiveness of decisions.

[0138] The embodiment of the present application further provides a product marketing device based on funnel analysis. It should be noted that the product marketing device based on funnel analysis can be used to execute the product marketing method based on funnel analysis provided by the embodiment of the present application. The device is used to realize the above-mentioned embodiment and preferred embodiment, and will not be described here. As used below, the term "module" can be a combination of software and / or hardware that realizes a predetermined function. Although the device described in the following embodiment is preferably realized in software, the realization of hardware or a combination of software and hardware is also possible and is conceived.

[0139] The product marketing device based on funnel analysis provided by the embodiment of the present application is introduced below.

[0140] Figure 4 is a schematic diagram of the product marketing device based on funnel analysis according to the embodiment of the present application. As shown in Figure 4 , the device includes an acquisition unit 10, a feature extraction unit 20, a clustering unit 30 and an analysis unit 40. The acquisition unit 10 is used to acquire a basic sample data set, the basic sample data set including customer data of a plurality of customers, the customer data including historical marketing activity data and customer information of the customers. The feature extraction unit 20 is used to extract a plurality of keywords from the basic sample data set, input all the keywords into a feature extraction model, obtain a plurality of customer group topics and corresponding keywords, and label each customer data with a topic label according to the keywords of the customer group topics. The clustering unit 30 is used to perform clustering analysis on all the customer data according to the topic labels of the customer data by using a clustering algorithm, and obtain a plurality of customer data sets, each of which includes a plurality of customer data. The analysis unit 40 is used to write the customer data of the customer data sets and the corresponding topic labels and keywords into a Doris database, analyze all the data in the Doris database by using funnel analysis to obtain an analysis result, and push a product to corresponding customers based on the analysis result.

[0141] The product marketing device based on funnel analysis provided in the present application comprises an acquisition unit, a feature extraction unit, a clustering unit and an analysis unit. The acquisition unit is configured to acquire a basic sample data set. The feature extraction unit is configured to extract a plurality of keywords from the basic sample data set, input all the keywords into a feature extraction model, obtain a plurality of customer group themes and keywords corresponding to each customer group theme, and label each customer data with a theme label according to all the keywords. The clustering unit is configured to perform clustering analysis on all the customer data according to the theme labels of the customer data by using a clustering algorithm, and obtain at least one customer data set and corresponding customer data. The analysis unit is configured to write all the data into a Doris database in batches, and query the data by using the Doris database to perform funnel analysis. The device combines the advantages of the feature extraction model and the clustering algorithm, orderly combines the input and output of the algorithm, and cross- validates the predicted results of the two algorithms, thereby greatly improving the accuracy of customer group positioning, and solving the problem of low data processing capacity of the funnel analysis tool in the prior art, which leads to delayed analysis results and inability to recommend accurate marketing products to customers.

[0142] In some embodiments, the feature extraction unit comprises a first processing module, a first construction module, a second processing module and a third processing module. The first processing module is configured to perform data cleaning and word segmentation processing on the data in the basic sample data set, and screen out keywords related to the business. The keywords at least include customer asset information, held product information, customer level data and the like. The keywords are one of Chinese characters, numbers and English. The first construction module is configured to construct a word segmentation dictionary, and count the number of occurrences of each keyword based on the word segmentation dictionary. The second processing module is configured to input the keywords and the corresponding number of occurrences into the feature extraction model to obtain model output data. The model output data at least includes the customer group theme, the keywords corresponding to the customer group theme, and the weight proportion of the keywords corresponding to the customer group theme. The third processing module is configured to label each customer data with a theme label according to the model output data, and determine the number of customers of each customer group theme according to the theme label, keywords and weight proportion of the keywords of each customer data. The feature extraction model is used to first perform data cleaning, word segmentation and screening of keywords related to the business on the massive customer information, construct a word segmentation dictionary and count the number of occurrences of the keywords.

[0143] In some embodiments, the second processing module includes a construction submodule and a processing submodule. The construction submodule is configured to construct the feature extraction model based on a Spark MLlib machine learning engine and obtain a preset number of keywords. The processing submodule is configured to input the keywords and corresponding occurrence times into the feature extraction model, and calculate the customer group theme, keywords corresponding to the customer group theme, and weight proportion of the keywords corresponding to the customer group theme based on the preset number of keywords using an Mallet version of the LDA algorithm to obtain the model output data. The customer behavior features of the massive original customer sample data in the HIVE library are extracted by the feature extraction model using the Spark MLlib platform.

[0144] In some embodiments, the clustering unit includes a conversion module and a training module. The conversion module is configured to convert the format of the customer data and corresponding theme labels into a target format to obtain a converted dataset. The target format is [keyword, keyword feature]. The training module is configured to iteratively train the converted dataset as input of a K-Means clustering algorithm to obtain an optimal number of customer data sets and customer data corresponding to each customer data set. Based on the massive amount of basic user data samples, it is found that the traditional clustering algorithm calculation process is very time-consuming. Therefore, the clustering algorithm uses a Mini Batch K-Means algorithm, which uses a part of the samples in the sample set to perform the traditional K-Means. This can avoid the calculation difficulty when the sample size is too large, and greatly speeds up the convergence speed of the algorithm.

[0145] In some embodiments, the analysis unit includes a first acquisition module and a batch writing module. The first acquisition module is configured to acquire the Doris database. An interface is configured to associate the original table and the target table through the Doris database. The original table is a data table including all customer data and corresponding theme labels and keywords. The target table is a query data table. The batch writing module is configured to batch write all data into the Doris database through a spark task, configure the customer data and corresponding theme labels and keywords into a Doris data engine to establish an index, and create a unique model with a customer unique number and other key fields as the primary key to obtain the Doris database after writing data. By building a self-developed Doris batch data synchronization framework, a visual task scheduling tool is used to batch write data into the Doris database through a spark task. The real-time query and calculation capabilities of Doris with high concurrency and low latency, and the powerful distributed capabilities supporting complex business analysis are used to load customer detail data and theme feature data into the Doris database, establish an index, and create a unique model to improve query efficiency.

[0146] In some embodiments, the analysis unit comprises a second acquisition module, a definition module and an association module, the second acquisition module is configured to acquire a target screening condition, the target screening condition comprises at least one screening logic and at least one sub-screening condition; the definition module is configured to define an index that needs to be aggregated and counted according to the target screening condition and a business requirement; and the association module is configured to perform an association query on the corresponding Doris data according to the index that needs to be aggregated and counted, and send the parsed data after the query to a target page. In the screening and statistical analysis scene of a large number of customers, the excellent performance and powerful distributed capability of the Doris OLAP data engine are the biggest advantages, compared with the traditional method, not only the response performance is greatly improved, but also the complex business analysis logic can be realized by using simple and easy-to-understand SQL (Structured Query Language).

[0147] In some embodiments, the acquisition unit comprises a determination module, a collection module and a fourth processing module, the determination module is configured to determine a target collection method, the target collection method comprises at least one of the following: mapping to a storage table based on a message queue, external heterogeneous data storage RPC direct reading collection, text file collection, real-time transaction API interface collection, HIVE data warehouse federation collection; the collection module is configured to collect initial sample data by using the target collection method, the initial sample data comprises real-time dynamic data and basic static data, the basic static data at least comprises customer information, and the real-time dynamic data at least comprises one of the following: asset details, user behavior data, real-time asset change situation; and the fourth processing module is configured to associate and summarize the basic static data and the real-time dynamic data by using a Spark preprocessing program to obtain a plurality of customer data, each of the customer data taking a customer number as a unique primary key. The above step collects mobile phone data from various marketing channels, and performs preliminary data cleaning, integration and preprocessing, thereby providing basic data for subsequent data analysis.

[0148] The product marketing device based on the funnel analysis comprises a processor and a memory, the acquisition unit and the like are stored in the memory as program units, and the corresponding functions are realized by the processor executing the program units stored in the memory. The modules are located in the same processor; or the modules are located in different processors in any combination.

[0149] The processor comprises a core, and the core retrieves the corresponding program unit from the memory. The core can be set to one or more, and the data processing capability of the funnel analysis tool in the prior art is low, which causes delay of the analysis result and inability to accurately recommend marketing products to customers, which can be solved by adjusting the core parameters.

[0150] The memory can include non-persistent memory in a computer readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash memory, in the form of at least one memory chip.

[0151] The embodiment of the present application provides a computer readable storage medium, the computer readable storage medium comprises a stored program, wherein the computer readable storage medium controls a device where the computer readable storage medium is located to execute the product marketing method based on funnel analysis when the program runs.

[0152] The embodiment of the present application provides a processor, the processor is used for running a program, wherein the product marketing method based on funnel analysis is executed when the program runs.

[0153] The embodiment of the present application provides an electronic device, the device comprises a processor, a memory and a program stored in the memory and capable of running on the processor, and the processor executes the program to realize at least the following steps:

[0154] Step S201, obtaining a basic sample data set, wherein the basic sample data set comprises customer data of a plurality of customers, and the customer data comprises historical marketing activity data and customer information of the customers;

[0155] Step S202, extracting a plurality of keywords from the basic sample data set, inputting all the keywords into a feature extraction model, obtaining a plurality of customer group themes and corresponding keywords, and labeling theme tags for the customer data according to the keywords of the customer group themes;

[0156] Step S203, performing clustering analysis on all the customer data according to the theme tags of the customer data by using a clustering algorithm, obtaining a plurality of customer data sets, and each customer data set comprising a plurality of customer data;

[0157] Step S204, writing the customer data of the customer data set and the corresponding theme tags and keywords into a Doris database, performing analysis on all the data in the Doris database by using funnel analysis to obtain an analysis result, and pushing a product to corresponding customers based on the analysis result.

[0158] The device in the present application can be a server, a PC, a PAD, a mobile phone and the like.

[0159] The present application also provides a computer program product, when executed on a data processing device, is suitable for executing a program initialized with at least the following method steps:

[0160] Step S201, obtain a basic sample data set, the basic sample data set includes customer data of a plurality of customers, and the customer data includes historical marketing activity data and customer information of the customers;

[0161] Step S202, extract a plurality of keywords from the basic sample data set, input all the keywords into a feature extraction model, obtain a plurality of customer group themes and corresponding keywords, and label theme tags for each customer data according to the keywords of the customer group themes;

[0162] Step S203, perform clustering analysis on all the customer data according to the theme tags of the customer data by using a clustering algorithm, obtain a plurality of customer data sets, and each customer data set includes a plurality of customer data;

[0163] Step S204, write the customer data of the customer data set and the corresponding theme tags and keywords into a Doris database, perform analysis on all the data in the Doris database by using funnel analysis to obtain an analysis result, and push a product to corresponding customers based on the analysis result.

[0164] Obviously, those skilled in the art should understand that each module or each step of the present application can be realized by a general computing device, which can be concentrated on a single computing device or distributed on a network composed of multiple computing devices, which can be realized by program codes executable by a computing device, so that they can be stored in a storage device and executed by a computing device, and in some cases, the steps shown or described can be executed in different order, or they can be manufactured into each integrated circuit module, or multiple modules or steps can be manufactured into a single integrated circuit module. Therefore, the present application is not limited to any specific hardware and software combination.

[0165] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can be in the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program codes.

[0166] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks.

[0167] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks.

[0168] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks.

[0169] In one typical configuration, the computing device includes one or more processors (CPU's), input / output interfaces, network interfaces, and memory.

[0170] The memory can include non-persistent memory and / or persistent memory, such as flash memory, read-only memory (ROM), and / or volatile or non-volatile random access memory (RAM), among others. The memory is an example of computer-readable media.

[0171] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to computing devices. According to the definition herein, computer-readable media does not include transitory media such as modulated data signals and carriers.

[0172] It should also be noted that the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, so that processes, methods, articles or devices including a series of elements not only include those elements, but also include other elements not explicitly listed or inherent to such processes, methods, articles or devices. Without more limitations, the element defined by the statement "including a" does not exclude the presence of additional identical elements in the process, method, article or device including the element.

[0173] From the above description, it can be seen that the above-mentioned embodiments of the present application achieve the following technical effects:

[0174] 1) The above-mentioned product marketing method based on funnel analysis of the present application first acquires a basic sample data set; extracts a plurality of keywords from the basic sample data set, and inputs all the keywords into a feature extraction model to obtain a plurality of customer group themes and keywords corresponding to each customer group theme, and labels each customer data with a theme label according to all the keywords; then a clustering algorithm is used to cluster all customer data according to the theme label of the customer data, to obtain at least one customer data set and corresponding customer data; finally, all data are written in batches to a Doris database, and the data are queried through the Doris database for funnel analysis. This method combines the advantages of feature extraction model and clustering algorithm, and orderly combines the input and output of the algorithm, and cross- validates the predicted results of the two algorithms, greatly improving the accuracy of customer group positioning, solving the problem that the data processing capacity of the funnel analysis tool in the prior art is low, resulting in delayed analysis results and inability to recommend accurate marketing products to customers.

[0175] 2), The product marketing device based on funnel analysis of the present application comprises an acquisition unit, a feature extraction unit, a clustering unit and an analysis unit, the acquisition unit is used for acquiring a basic sample data set; the feature extraction unit is used for extracting a plurality of keywords from the basic sample data set, inputting all the keywords into a feature extraction model, obtaining a plurality of customer group themes and keywords corresponding to each customer group theme, and labeling each customer data with a theme label according to all the keywords; the clustering unit is used for clustering all customer data according to the theme label of the customer data by using a clustering algorithm to obtain at least one customer data set and corresponding customer data; the analysis unit is used for batch writing all data into a Doris database, and querying the data through the Doris database to perform funnel analysis. The device combines the advantages of the feature extraction model and the clustering algorithm, orderly combines the input and output of the algorithm, and cross- validates the predicted results of the two algorithms, greatly improving the accuracy of customer group positioning of the present application, solving the problem of low data processing capacity of the funnel analysis tool in the prior art, leading to delayed analysis results, and unable to recommend accurate marketing products to customers.

[0176] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Those skilled in the art can make various changes and modifications to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A product marketing method based on funnel analysis, characterized in that, include: Obtain a basic sample dataset, which includes customer data for multiple customers, including their historical marketing activity data and customer information; Multiple keywords are extracted from the basic sample dataset, and all the keywords are input into the feature extraction model to obtain multiple customer group themes and corresponding keywords. Based on the keywords of the customer group themes, the customer data is labeled with theme tags. A clustering algorithm is used to perform cluster analysis on all the customer data based on the topic tags of the customer data, resulting in multiple customer data sets, each of which includes multiple customer data sets; The customer data, along with the corresponding topic tags and keywords, are written into the Doris database. Funnel analysis is used to analyze all the data in the Doris database to obtain analysis results, and products are pushed to the corresponding customers based on the analysis results.

2. The method according to claim 1, characterized in that, Multiple keywords are extracted from the basic sample dataset, and all keywords are input into a feature extraction model to obtain multiple customer group themes and corresponding keywords. Based on the keywords of the customer group themes, topic tags are assigned to each customer data point, including: The data in the basic sample dataset is cleaned and segmented, and keywords related to the business are selected. The keywords include at least the following dimensions: customer asset information, product holding information, customer level, etc. The keywords can be Chinese characters, numbers, or English words. Construct a word segmentation dictionary, and count the occurrence frequency of each keyword based on the word segmentation dictionary; The keywords and their corresponding frequency of occurrence are input into the feature extraction model to obtain model output data. The model output data includes at least the customer group theme, the keywords corresponding to the customer group theme, and the weight percentage of the keywords corresponding to the customer group theme. The model output data is used to label each customer data with topic tags, and the number of customers for each customer group topic is determined based on the topic tags, keywords, and the weight ratio of the keywords for each customer data.

3. The method according to claim 2, characterized in that, The keywords and their corresponding frequency of occurrence are input into the feature extraction model to obtain the model output data, including: The feature extraction model is built based on the Spark MLlib machine learning engine, and a preset number of keywords is obtained; The keywords and their corresponding occurrence counts are input into the feature extraction model, and the Mallet version of the LDA algorithm is used to calculate the customer group theme, the keywords corresponding to the customer group theme, and the weight ratio of the keywords corresponding to the customer group theme based on the preset number of keywords, so as to obtain the model output data.

4. The method according to claim 1, characterized in that, Clustering algorithms are used to perform cluster analysis on all the customer data based on the topic tags of the customer data, resulting in multiple customer data sets. Each customer data set includes multiple sets of customer data, including: The customer data and corresponding topic tags are converted into the target format to obtain the converted dataset; The transformed dataset is used as input to the K-Means clustering algorithm for iterative training to obtain the optimal number of sets of customer data and the customer data of each customer data set.

5. The method according to claim 1, characterized in that, Write all data to the Doris database in batches, including: Obtain the Doris database, wherein an interface is configured to associate the original table and the target table through the Doris database, wherein the original table is a data table including all the customer data and the corresponding topic tags and keywords, and the target table is a query data table; All data is written to the Doris database in batches using a Spark task. The customer data, along with the corresponding topic tags and keywords, are then indexed in the Doris data engine. A unique model is created, using the customer's unique ID and other key fields as the primary key, resulting in the Doris database after the data has been written.

6. The method according to claim 1, characterized in that, Funnel analysis was used to analyze all data in the Doris database to obtain the following results: Obtain target filtering conditions, which consist of at least one filtering logic and at least one sub-filtering condition; Based on the target filtering conditions and business needs, define the indicators that need to be aggregated and statistically analyzed; Based on the indicators that need to be aggregated and statistically analyzed, perform related queries on the corresponding Doris data, parse the queried data, and send it to the target page.

7. The method according to claim 1, characterized in that, Obtain the basic sample dataset, including: Determine the target data collection method, which includes at least one of the following: message queue mapping to storage table, direct RPC reading from external heterogeneous data storage, text file collection, real-time transaction API interface collection, and HIVE data warehouse federated collection. Initial sample data is collected using the target acquisition method described above. The initial sample data includes real-time dynamic data and basic static data. The basic static data includes at least customer information, and the real-time dynamic data includes at least one of the following: asset details, user behavior data, and real-time asset changes. The basic static data and the real-time dynamic data are associated and aggregated using Spark preprocessing to obtain multiple customer data entries, each with the customer ID as a unique primary key.

8. A product marketing device based on funnel analysis, characterized in that, include: The acquisition unit is used to acquire a basic sample dataset, which includes customer data of multiple customers, including the customers' historical marketing activity data and customer information. The feature extraction unit is used to extract multiple keywords from the basic sample dataset, input all the keywords into the feature extraction model to obtain multiple customer group themes and corresponding keywords, and label each customer data with theme tags according to the keywords of the customer group themes; A clustering unit is used to perform clustering analysis on all the customer data according to the topic tags of the customer data using a clustering algorithm to obtain multiple customer data sets, each of the customer data sets including multiple customer data sets. The analysis unit is used to write customer data from the customer data set, along with the corresponding topic tags and keywords, into the Doris database, perform funnel analysis on all data in the Doris database to obtain analysis results, and push products to the corresponding customers based on the analysis results.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform the product marketing method based on funnel analysis as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including a product marketing method based on funnel analysis as described in any one of claims 1 to 7.

Images