Product precision marketing method based on dimension classification

Abstract

The present application relates to the field of marketing recommendation technology, and in particular to a method for precision product marketing based on dimensional grading, including: orderly sample clustering of the heat sequence of each product within a preset time period to obtain a variety of historical marketing environments; in the purchase records of target users, with the goal of minimizing the prediction error between the predicted recommendation degree of each product and the actual purchased product, obtaining the priority of each analysis dimension of the target user in each historical marketing environment; calculating the real-time priority of each analysis dimension of the target user based on the similarity between the real-time marketing environment and each historical marketing environment; calculating the sum of the product of the real-time attribute value and the real-time priority of each analysis dimension to obtain the real-time recommendation degree of each product, and making product recommendations according to the real-time recommendation degree. The technical solution of the present application can make personalized recommendations for target users and realize precision marketing of products.

Description

Technical Field

[0001] The present application relates to the field of marketing recommendation technology, and in particular to a product precision marketing method based on dimensional grading. Background Art

[0002] With the rapid development of the Internet of Things and logistics technology, online shopping malls have become the first choice for many users. In order to enhance brand awareness, attract potential customers, and promote sales, merchants need to adopt a series of marketing methods in online shopping malls to recommend products to users in order to achieve precision marketing.

[0003] At present, a patent application document with publication number CN118172091A discloses an online marketing strategy recommendation method and system, which includes: collecting consumption records and operation data, preprocessing the collected consumption records and operation data, and fusing the processed data consumption records, operation data and timestamps to form a standard data set; the consumption records include consumption amounts, consumption frequencies and commodity types; the operation data include user clicks and browses data, monitors visited pages and browsing time; the timestamp is the time information corresponding to the consumption records and operation data; receiving the standard data set, extracting data features from the standard data set, using a clustering algorithm based on the data features to divide consumers into different groups according to the weight ratio of different factors, and performing a detailed feature description for each group; the number of categories of the group can be adjusted by the weight ratio of different factors; receiving the group division results and feature descriptions, formulating corresponding marketing strategies, and pushing suitable commodities.

[0004] The above method divides all consumers into multiple groups based on a standard data set consisting of operation data and consumption records, and formulates marketing strategies based on the group division results and feature descriptions, and then pushes suitable products. However, the group division results cannot accurately describe the characteristics of each user, and thus cannot make personalized recommendations for each user, and cannot achieve precise marketing of products. Summary of the invention

[0005] In order to solve the technical problem of being unable to achieve precise product marketing, this application provides a precise product marketing method based on dimensional grading, which can make personalized recommendations for target users and achieve precise product marketing.

[0006] The present application provides a product precision marketing method based on dimension grading, the method comprising: performing ordered sample clustering on the heat sequence of each product within a preset time period to obtain multiple historical marketing environments, the heat being product sales volume or product praise rate; in the purchase record of the target user, taking minimizing the prediction error between the predicted recommendation degree of each product and the actual purchased product as the goal, using an optimization algorithm to obtain the priority of each analysis dimension of the target user in each historical marketing environment, the predicted recommendation degree of the product being the sum of the product of the priority of each analysis dimension and the attribute value, the analysis dimensions including the product type, the last purchase interval and the last purchase amount; calculating the similarity between the real-time marketing environment and each historical marketing environment, taking the historical marketing environment with a similarity greater than a similarity threshold as a candidate marketing environment, performing weighted summation on the priority of the target user in each candidate marketing environment to obtain the real-time priority of each analysis dimension of the target user; calculating the sum of the product of the real-time attribute value and the real-time priority of each analysis dimension to obtain the real-time recommendation degree of each product, and recommending products in order of the real-time recommendation degree from large to small.

[0007] Orderly sample clustering is performed on the heat sequences of all products to obtain multiple clusters. A cluster includes the heat subsequences of all products, which can reflect the relative relationship between the sales of each product. A cluster is regarded as a historical marketing environment. According to the purchase records of target users, the priority of each analysis dimension of the target user in each historical marketing environment is determined with the goal of minimizing the prediction error between the predicted recommendation degree of each product and the actual purchased product. The priority of each analysis dimension can realize personalized recommendation for the target user in the corresponding historical marketing environment. According to the similarity between the real-time marketing environment and each historical marketing environment, the priority of the target user in each candidate marketing environment is weighted and summed to obtain the real-time priority of each analysis dimension of the target user. The real-time priority can realize personalized recommendation for the target user in the real-time marketing environment and realize precision marketing.

[0008] Preferably, obtaining a plurality of historical marketing environments comprises: performing ordered sample clustering of the heat sequence of each product according to an initial clustering number to obtain a plurality of cluster clusters, and calculating an evaluation value of the initial clustering number, wherein the evaluation value is negatively correlated with the average variance of each heat subsequence in the cluster cluster and the cluster similarity of adjacent cluster clusters; updating the initial clustering number multiple times, and drawing an evaluation value curve of each initial clustering number; taking the initial clustering number corresponding to the inflection point in the evaluation value curve as the target clustering number, and the plurality of cluster clusters corresponding to the target clustering number correspond to a plurality of historical marketing environments.

[0009] The heat sequence of each product can reflect the relative relationship between the sales of all products, continuously adjust the initial number of clusters, balance the clustering effect and the number of historical marketing environments, so that the heat of each product in each historical marketing environment remains basically stable, thereby accurately quantifying the marketing environment.

[0010] Preferably, the cluster similarity is calculated by: calculating the similarity of each product according to the DTW distance of the heat subsequences in any two clusters; and taking the average value of the similarities of each product as the cluster similarity of the any two clusters.

[0011] Preferably, the cluster similarity is calculated by: calculating the average value of each heat subsequence in any cluster to obtain the cluster center; calculating the cluster similarity of any two clusters based on the Euclidean distance of the cluster centers, and the cluster similarity is negatively correlated with the Euclidean distance of the cluster centers.

[0012] Preferably, using an optimization algorithm to obtain the priority of each analysis dimension of a target user in each historical marketing environment includes: obtaining a purchase record of a target user in a historical marketing environment, the purchase record including the attribute values ​​of each analysis dimension of all products at the time of purchase and the actually purchased products; initializing the priority of each analysis dimension, calculating the predicted recommendation degree of each product at the time of purchase based on the attribute values ​​of each analysis dimension of all products in the purchase record, and setting the recommendation degree label of the actually purchased product to 1, and setting the recommendation degree labels of other products other than the actually purchased product to 0 to calculate the prediction error; using an optimization algorithm to adjust the priority of each analysis dimension multiple times until the value of the prediction error reaches a minimum, thereby obtaining the priority of each analysis dimension of the target user in the historical marketing environment.

[0013] The target users have different priorities in analysis dimensions under different marketing environments. For any historical marketing environment, the purchase records of the target users under that historical marketing environment are obtained, and the predicted recommendation degrees of each product at the time of purchase are calculated based on the purchase records. The priority of each analysis dimension is adjusted based on the prediction error between the predicted recommendation degree and the actual purchased product. When the prediction error reaches the minimum value, the priority of each analysis dimension of the target users under that historical marketing environment is obtained. Based on the priority of each analysis dimension, personalized recommendations for the target users under that historical marketing environment can be achieved.

[0014] Preferably, the prediction error satisfy: , For the historical marketing environment The number of target users' purchase records. is the number of products, For purchase records Medium Products The predicted recommendation degree, For purchase records Medium Products The recommended label.

[0015] A calculation method for prediction error is given. The smaller the prediction error is, the closer the predicted recommendation degree of the actually purchased product is to 1. The closer the predicted recommendation degree of products other than the actually purchased product is to 0, the more accurate the priority of each analysis dimension is.

[0016] Preferably, the real-time marketing environment includes the popularity of each product in the current collection period; the real-time marketing environment and the historical marketing environment Similarity for: , For real-time marketing environment, For the historical marketing environment The cluster center of .

[0017] Preferably, target user analysis dimensions Real-time priority satisfy: , For the The similarity between the candidate marketing environment and the real-time marketing environment, is the sum of the similarities between all candidate marketing environments and the real-time marketing environment, For the target users Analytical Dimensions in Candidate Marketing Environments priority.

[0018] Preferably, after obtaining the real-time recommendation degree of each product, the method further comprises: Calculate the personalization coefficient , the personalized coefficient satisfy: ; Based on the personalization coefficient, the real-time recommendation degree is corrected, and the corrected real-time recommendation degree satisfy: , For real-time marketing environment products The heat, For the product before correction Real-time recommendation.

[0019] Taking into account the special case that the number of purchase records in the historical marketing environment is 0, the personalization coefficient of the target user is determined according to the amount of data that can be collected from the target user in the candidate marketing environment. The personalization coefficient is combined with the degree of preference of all users for the product in the real-time marketing environment and the personalized recommendation results of the target user to obtain the final recommendation result and realize product precision marketing.

[0020] The technical solution of this application has the following beneficial technical effects:

[0021] The above-mentioned product precision marketing method based on dimension grading provided in the embodiment of the present application performs ordered sample clustering on the heat sequences of all products to obtain multiple cluster clusters, wherein a cluster cluster includes the heat subsequences of all products, which can reflect the relative relationship of the sales conditions of each product, and regards a cluster cluster as a historical marketing environment; according to the purchase record of the target user, the priority of each analysis dimension of the target user in each historical marketing environment is determined with the goal of minimizing the prediction error between the predicted recommendation degree of each product and the actual purchased product, and the priority of each analysis dimension can realize personalized recommendation for the target user in the corresponding historical marketing environment; further, according to the similarity between the real-time marketing environment and each historical marketing environment, the priority of the target user in each candidate marketing environment is weighted and summed to obtain the real-time priority of each analysis dimension of the target user; according to the real-time priority of each analysis dimension, the real-time recommendation degree of each product is calculated, and the product recommendation is performed in the order of the real-time recommendation degree from large to small, so as to realize personalized recommendation for the target user, and then realize product precision marketing. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] By reading the detailed description below with reference to the accompanying drawings, the above and other purposes, features and advantages of the exemplary embodiments of the present application will become easy to understand. In the accompanying drawings, several embodiments of the present application are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0023] Figure 1 It is a flowchart of a product precision marketing method based on dimensional grading according to an embodiment of the present application. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0025] It should be understood that when the application uses the terms "first", "second", etc., they are only used to distinguish different objects, rather than to describe a specific order. The terms "include" and "comprise" used in the application indicate the presence of the described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or their collections.

[0026] The present application provides a product precision marketing method based on dimension grading, which is used in the client of an online mall. When a target user logs in to the online mall on the client, recommended products will be displayed on the client interface. The target user can enter the purchase interface of the recommended product by clicking on the recommended product.

[0027] It should be noted that the online mall can also be an online store of any enterprise or any brand, and this application does not impose any restrictions.

[0028] See also Figure 1 , is a flow chart of a product precision marketing method based on dimensional grading according to an embodiment of the present application. According to different requirements, the order of the steps in the flow chart can be changed, and some steps can be omitted.

[0029] S11, performing ordered sample clustering on the heat sequence of each product within a preset time period to obtain a variety of historical marketing environments, wherein the heat is product sales volume or product praise rate.

[0030] In one embodiment, a product list of an online mall is obtained, and the product list includes all products sold; the popularity of any product in multiple collection cycles in a preset time period is collected to obtain a popularity sequence of the product, and the popularity can reflect the degree of liking of the product by all users in the collection cycle. The popularity is the product sales volume or the product praise rate. The preset time period is the past year, and the collection cycle is 1 day or 12 hours, that is, the preset time period includes multiple collection cycles, wherein the product sales volume is the sales quantity of the product in one collection cycle, and the product praise rate is the ratio of the number of praises to the sales quantity of the product in one collection cycle.

[0031] The heat sequence of all products can reflect the relative relationship of the sales of each product, and then reflect the changes in the marketing environment. For example, the sales volume of product 1 is much greater than that of product 2, which indicates that in the current marketing environment, users are more inclined to buy product 2. Therefore, by clustering the heat sequences of all products in an ordered sample, multiple clusters can be obtained, and a cluster includes the heat subsequences of all products. A cluster can be regarded as a historical marketing environment, which can reflect the relative relationship of the sales of each product in the corresponding time period.

[0032] Specifically, obtaining a variety of historical marketing environments includes: performing ordered sample clustering on the heat sequence of each product according to an initial clustering number to obtain a plurality of cluster clusters, and calculating an evaluation value of the initial clustering number, wherein the evaluation value is negatively correlated with the average variance of each heat subsequence in the cluster cluster and the cluster similarity of adjacent cluster clusters; updating the initial clustering number multiple times, and drawing an evaluation value curve for each initial clustering number; taking the initial clustering number corresponding to the inflection point in the evaluation value curve as the target clustering number, and the plurality of cluster clusters corresponding to the target clustering number correspond to a variety of historical marketing environments.

[0033] Among them, the value of the initial cluster number is 2. The ordered sample clustering divides the heat sequences of all products into multiple heat subsequences by continuously searching for the optimal split point. The heat subsequences of all products between adjacent optimal split points are regarded as a cluster, so that the average variance of the heat subsequences within a cluster is minimized, and the difference of the heat subsequences between adjacent clusters is maximized.

[0034] Specifically, in the process of clustering the heat sequence of each product in an ordered sample according to the initial cluster number, it is necessary to define the diameter of the cluster. The diameter of a cluster is equal to the average variance of all heat subsequences in the cluster. The diameter of the clusters satisfy:

[0035] , and Respectively The starting and ending collection periods of the heat subsequences in the clusters, For the Cluster products The variance of the heat subsequence, is the number of all products. After defining the diameter of the cluster, the ordered sample clustering can find the optimal split point in the heat sequence, and divide the heat sequence of each product into several clusters of the initial clustering. One cluster includes the heat subsequences of all products. The implementation method of ordered sample clustering is a well-known technology for those skilled in the art and will not be repeated here.

[0036] Exemplarily, if the number of products is 10 and the heat sequence includes 365 collection cycles, the heat sequence of all products can be regarded as time series data with 10 rows and 365 columns, where each row corresponds to a product; when the initial number of clusters is 2, and the optimal split point found by ordered sample clustering is the 59th collection cycle, the heat subsequences of all products between the 1st collection cycle and the 59th collection cycle constitute a cluster cluster, and the heat subsequences of all products between the 60th collection cycle and the 365th collection cycle constitute another cluster cluster, and the clustering result is obtained when the initial number of clusters is 2. It can be understood that when the initial number of clusters is 3, 2 optimal splits will be found, thereby dividing the heat sequences of all products into 3 cluster clusters.

[0037] In one embodiment, the initial number of clusters correspond Clusters, initial number of clusters Evaluation value for: , For the The average variance of each heat subsequence in the clusters, For the Clusters and The cluster similarity of the clusters.

[0038] The cluster similarity is calculated by: calculating the similarity of each product based on the DTW distance of the heat subsequences in any two clusters; and taking the average value of the similarities of each product as the cluster similarity of the any two clusters.

[0039] Specifically, no. Clusters and Cluster products Similarity for: , For the Clusters and Cluster products DTW distance of heat subsequence, is an exponential function with base e.

[0040] In another embodiment, the cluster similarity is calculated by: calculating the average value of each heat subsequence in any cluster to obtain the cluster center; calculating the cluster similarity of any two clusters based on the Euclidean distance of the cluster centers, and the cluster similarity is negatively correlated with the Euclidean distance of the cluster centers.

[0041] Specifically, no. Clusters and The cluster similarity of clusters satisfy: , and Respectively Clusters and The cluster centers of the clusters, is an exponential function with base e, express and The Euclidean distance of .

[0042] In this way, the heat sequences of each product in a preset time period are clustered according to ordered sample clustering, and multiple optimal segmentation points are determined, thereby obtaining multiple clusters. A cluster includes the heat subsequences of all products and can reflect the sales of all products. A cluster is regarded as a historical marketing environment.

[0043] S12, in the purchase records of target users, with the goal of minimizing the prediction error between the predicted recommendation degree of each product and the actual purchased product, the optimization algorithm is used to obtain the priority of each analysis dimension of the target user in each historical marketing environment. The predicted recommendation degree of the product is the sum of the product of the priority of each analysis dimension and the attribute value. The analysis dimensions include product type, last purchase interval and last purchase amount.

[0044] In one embodiment, the target user is any user of the online mall. The sales of each product in different historical marketing environments are different, which will lead to different priorities of each analysis dimension when the target user purchases the product. Therefore, it is necessary to use the historical purchase record of the target user within a preset time period to determine the priority of each analysis dimension of the target user in each historical marketing environment.

[0045] The analysis dimensions include product type, last purchase interval, and last purchase amount; in other embodiments, the analysis dimensions also include product price and product delivery distance; the predicted recommendation degree of each product can be obtained by using the attribute values ​​of the analysis dimensions and the priority of each analysis dimension. The predicted recommendation satisfy: , is the number of analysis dimensions, For analysis dimension The priority of For products In the analysis dimension The attribute value of .

[0046] In one embodiment, obtaining the priority of each analysis dimension of a target user in each historical marketing environment using an optimization algorithm includes: obtaining a purchase record of a target user in a historical marketing environment, the purchase record including the attribute values ​​of each analysis dimension of all products at the time of purchase and the actually purchased products; initializing the priority of each analysis dimension, calculating the predicted recommendation degree of each product at the time of purchase based on the attribute values ​​of each analysis dimension of all products in the purchase record, and setting the recommendation degree label of the actually purchased product to 1, and setting the recommendation degree labels of other products other than the actually purchased product to 0 to calculate the prediction error; using an optimization algorithm to adjust the priority of each analysis dimension multiple times until the value of the prediction error reaches a minimum, thereby obtaining the priority of each analysis dimension of the target user in the historical marketing environment.

[0047] The number of purchase records is not less than 1, and the prediction error satisfy:

[0048] , For the historical marketing environment The number of target users' purchase records. is the number of products, For purchase records Medium Products The predicted recommendation degree, For purchase records Medium Products The recommendation label of In the purchase records, in response to the product To actually purchase the product, The recommendation label of the product is 1, otherwise, The recommendation label is 0.

[0049] Among them, the optimization algorithm is a simulated annealing algorithm, a genetic algorithm or a gradient descent method, which is not limited in this application.

[0050] Since it is impossible to ensure that the purchase records of target users can be collected in each historical marketing environment, there will be a situation where the number of purchase records in a historical marketing environment is 0; if the purchase records cannot be collected, personalized recommendations for target users in that historical marketing environment cannot be achieved. Therefore, the priority of the target user in each analysis dimension in that historical marketing environment is directly set to 0.

[0051] In this way, the priority of each analysis dimension of the target user in each historical marketing environment is determined based on the target user's purchase record. The priority of each analysis dimension can be used to accurately calculate the predicted recommendation degree of each product, thereby realizing personalized recommendations for the target user in the historical marketing environment.

[0052] S13, calculate the similarity between the real-time marketing environment and each historical marketing environment, take the historical marketing environment with a similarity greater than the similarity threshold as the candidate marketing environment, perform weighted summation on the priority of the target user in each candidate marketing environment, and obtain the real-time priority of each analysis dimension of the target user.

[0053] In one embodiment, the real-time marketing environment is the popularity of each product in the current collection period. The number of historical marketing environments is equal to the number of target clusters, and one historical marketing environment corresponds to one cluster center. Similarity for: , For real-time marketing environment, For the historical marketing environment The cluster center of is an exponential function with base e.

[0054] In one embodiment, the target user analysis dimension Real-time priority satisfy:

[0055] , For the The similarity between the candidate marketing environment and the real-time marketing environment, is the sum of the similarities between all candidate marketing environments and the real-time marketing environment, For the target users Analytical Dimensions in Candidate Marketing Environments The similarity threshold is set to 0.6.

[0056] In this way, the real-time priority of each analysis dimension of the target user in the real-time marketing environment is obtained, and the real-time priority of each analysis dimension can be used to achieve personalized recommendations for the target user in the real-time marketing environment.

[0057] S14, calculating the sum of the products of the real-time attribute value and the real-time priority of each analysis dimension, obtaining the real-time recommendation degree of each product, and making product recommendations in descending order of the real-time recommendation degree.

[0058] In one embodiment, the real-time attribute values ​​of each analysis dimension of any product at the current moment are collected, and the sum of the products of the real-time attribute values ​​of each analysis dimension and the real-time priority is calculated to obtain the real-time recommendation degree of any product.

[0059] Specifically, the product Real-time recommendation satisfy: , is the number of analysis dimensions, For target users Real-time priority of each analysis dimension, For products No. After obtaining the real-time attribute values ​​of all products, we can recommend products in descending order based on the real-time recommendation degree to achieve precise product marketing.

[0060] In another embodiment, due to the special case that the number of purchase records in the historical marketing environment is 0, the priority of each analysis dimension of the target user in the historical marketing environment is 0. If the target user is a new user of the online mall or the target user has a small amount of data, the recommendation result will be inaccurate. Therefore, after obtaining the real-time recommendation degree of each product, the method further includes: according to the total number of purchase records of the target user in each candidate marketing environment Calculate the personalization coefficient , the personalized coefficient satisfy: ; Based on the personalization coefficient, the real-time recommendation degree is corrected, and the corrected real-time recommendation degree satisfy: , For real-time marketing environment products The heat, For the product before correction Real-time recommendation.

[0061] In this way, the personalization coefficient of the target user is determined based on the amount of data that can be collected from the target user in the candidate marketing environment. The personalization coefficient is combined with the degree of preference of all users for the product in the real-time marketing environment and the personalized recommendation results of the target user to obtain the final recommendation result and achieve precise product marketing.

[0062] The above-mentioned product precision marketing method based on dimension grading provided in the embodiment of the present application performs ordered sample clustering on the heat sequences of all products to obtain multiple cluster clusters, wherein a cluster cluster includes the heat subsequences of all products, which can reflect the relative relationship of the sales conditions of each product, and regards a cluster cluster as a historical marketing environment; according to the purchase record of the target user, the priority of each analysis dimension of the target user in each historical marketing environment is determined with the goal of minimizing the prediction error between the predicted recommendation degree of each product and the actual purchased product, and the priority of each analysis dimension can realize personalized recommendation for the target user in the corresponding historical marketing environment; further, according to the similarity between the real-time marketing environment and each historical marketing environment, the priority of the target user in each candidate marketing environment is weighted and summed to obtain the real-time priority of each analysis dimension of the target user; according to the real-time priority of each analysis dimension, the real-time recommendation degree of each product is calculated, and the product recommendation is performed in the order of the real-time recommendation degree from large to small, so as to realize personalized recommendation for the target user, and then realize product precision marketing.

[0063] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0064] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent application. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application.

Claims

1. A product precision marketing method based on dimension grading, characterized in that: The method comprises: Get a variety of historical marketing environments, including: According to the initial clustering number, the heat sequence of each product is clustered in an ordered sample manner to obtain multiple clusters, and the evaluation value of the initial clustering number is calculated. The evaluation value is negatively correlated with the average variance of each heat subsequence in the cluster and the cluster similarity of adjacent clusters. Update the initial number of clusters multiple times and draw the evaluation value curve of each initial number of clusters; The initial cluster number corresponding to the inflection point in the evaluation value curve is used as the target cluster number, and the multiple clusters corresponding to the target cluster number correspond to multiple historical marketing environments; The popularity refers to product sales or product praise rate; In the target user's purchase record, the goal is to minimize the prediction error between the predicted recommendation degree of each product and the actual purchased product. The optimization algorithm is used to obtain the priority of each analysis dimension of the target user in each historical marketing environment. The predicted recommendation degree of the product is the sum of the product of the priority of each analysis dimension and the attribute value. The analysis dimensions include product type, last purchase interval and last purchase amount. Calculate the similarity between the real-time marketing environment and each historical marketing environment, take the historical marketing environment with a similarity greater than the similarity threshold as the candidate marketing environment, perform weighted summation on the priority of the target user in each candidate marketing environment, and obtain the real-time priority of each analysis dimension of the target user; The sum of the products of the real-time attribute value and the real-time priority of each analysis dimension is calculated to obtain the real-time recommendation degree of each product, and product recommendations are made in descending order of real-time recommendation degree.

2. The product precision marketing method based on dimensional grading according to claim 1, characterized in that: The cluster similarity is calculated as follows: The similarity of each product is calculated based on the DTW distance of the heat subsequences in any two clusters; the average value of the similarity of each product is taken as the cluster similarity of the any two clusters.

3. The product precision marketing method based on dimensional grading according to claim 1, characterized in that: The cluster similarity is calculated as follows: The average value of each heat subsequence in any cluster is calculated to obtain the cluster center; the cluster similarity between any two clusters is calculated based on the Euclidean distance of the cluster centers, and the cluster similarity is negatively correlated with the Euclidean distance of the cluster centers.

4. The product precision marketing method based on dimensional grading according to claim 1, characterized in that: The optimization algorithm is used to obtain the priority of each analysis dimension of the target user in each historical marketing environment, including: Obtaining a purchase record of a target user in a historical marketing environment, wherein the purchase record includes attribute values ​​of all products in each analysis dimension at the time of purchase and the actual purchased products; Initialize the priority of each analysis dimension, calculate the predicted recommendation degree of each product at the time of purchase based on the attribute values ​​of each analysis dimension of all products in the purchase record, and set the recommendation degree label of the actually purchased product to 1, and the recommendation degree labels of other products other than the actually purchased product to 0 to calculate the prediction error; The priority of each analysis dimension is adjusted multiple times using an optimization algorithm until the value of the prediction error reaches a minimum, thereby obtaining the priority of each analysis dimension of the target user in the historical marketing environment.

5. The product precision marketing method based on dimensional grading according to claim 4, characterized in that: The prediction error satisfy: , For the historical marketing environment The number of target users' purchase records. is the number of products, For purchase records Medium Products The predicted recommendation degree, For purchase records Medium Products The recommended label.

6. The product precision marketing method based on dimensional grading according to claim 1, characterized in that: The real-time marketing environment includes the popularity of each product in the current collection period; Real-time marketing environment vs. historical marketing environment Similarity for: , For real-time marketing environment, For the historical marketing environment The cluster center of .

7. The product precision marketing method based on dimensional grading according to claim 6, characterized in that: Target user analysis dimensions Real-time priority satisfy: , For the The similarity between the candidate marketing environment and the real-time marketing environment, is the sum of the similarities between all candidate marketing environments and the real-time marketing environment, For the target users Analytical Dimensions in Candidate Marketing Environments priority.

8. The product precision marketing method based on dimensional grading according to claim 1, characterized in that: After obtaining the real-time recommendation degree of each product, the method further includes: Based on the total number of purchase records of target users in each candidate marketing environment Calculate the personalization coefficient , the personalized coefficient satisfy: ; The real-time recommendation degree is modified based on the personalization coefficient, and the modified real-time recommendation degree satisfy: , For real-time marketing environment products The heat, For the product before correction Real-time recommendation.

Images