Big data private domain accurate user pushing method

Abstract

The invention relates to a big data private domain accurate user pushing method, and relates to the technical field of data processing, and the method comprises the steps: determining a plurality of commodity classes based on the commodity feature information of a plurality of on-sale commodities; determining a plurality of test users corresponding to the commodity classes based on the shopping behavior data of the plurality of test users on the e-commerce platform, and calculating commodity class related parameters of the two commodity classes; determining a plurality of user clusters according to the user portraits of the plurality of test users and the shopping behavior data on the e-commerce platform; based on the shopping behavior data of the to-be-analyzed user on the e-commerce platform and the plurality of user clusters, establishing a user portrait of the to-be-analyzed user, and combining commodity features of at least one promotion commodity corresponding to each piece of to-be-pushed preferential information, a plurality of commodity classes and commodity class related parameters of any two commodity classes, and determining the to-be-pushed commodity and / or the to-be-pushed preferential information matched with the to-be-analyzed user. The method has the advantage of accurately pushing the adaptive commodity or preferential information related to the user.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method for accurately pushing data to private domain users. Background Art

[0002] With the development of information technology, people can now buy products from all over the country without leaving their homes. They only need to search for their favorite products on shopping apps and pay online. The merchants will then send the products to the users through the express delivery industry, greatly facilitating the users' shopping experience.

[0003] However, most current shopping software can only push related products based on the user's search results. There may be a large number of merchants selling the same product, and some merchants may be offering product discounts. However, shopping software cannot effectively filter out products that are offering discounts, and users need to search on their own, which is very inconvenient.

[0004] Therefore, it is necessary to provide a big data private domain precise user push method for accurately pushing user-related adaptive products or preferential information. Summary of the invention

[0005] The present invention provides a method for accurately pushing users in a private domain with big data, including: obtaining commodity feature information of a plurality of commodities on sale on an e-commerce platform; classifying the plurality of commodities on sale based on the commodity feature information of the plurality of commodities on sale to determine a plurality of commodity categories; obtaining user portraits of a plurality of test users and shopping behavior data on the e-commerce platform, wherein the shopping behavior data at least includes order data, shopping cart data and commodity retrieval data, and the user portraits at least include occupational attributes, age attributes, family status attributes and preference attributes; for each of the commodity categories, based on the shopping behavior data of the plurality of test users on the e-commerce platform, determining a plurality of test users corresponding to the commodity category; for any two of the commodity categories, based on the plurality of test users corresponding to each of the commodity categories, calculating commodity category-related parameters of the two commodity categories; and calculating the commodity category-related parameters of the plurality of test users according to the plurality of test users. Based on the user portrait and the shopping behavior data on the e-commerce platform, the multiple test users are clustered to determine multiple user clusters; the shopping behavior data of at least one user to be analyzed on the e-commerce platform is obtained; for each user to be analyzed, a user portrait of the user to be analyzed is established based on the shopping behavior data of the user to be analyzed on the e-commerce platform and the multiple user clusters; at least one piece of discount information to be pushed is obtained; for each piece of discount information to be pushed, the discount information to be pushed is analyzed to determine the product features of at least one promotional product corresponding to the discount information to be pushed; based on the user portrait of the user to be analyzed, the product features of at least one promotional product corresponding to each piece of discount information to be pushed, multiple product categories and product category-related parameters of any two product categories, the pushed products and / or discount information to be pushed that match the user to be analyzed are determined.

[0006] Furthermore, based on the shopping behavior data of the multiple test users on the e-commerce platform, multiple test users corresponding to the commodity category are determined, including: for each of the test users, the commodity retrieval data of the test user is screened to determine valid commodity retrieval data; for each of the test users, based on the order data of the test user, the number of commodity purchases by the test user in multiple historical time periods is determined, based on the shopping cart data of the test user, the number of additional purchases of the commodity by the test user in multiple historical time periods is determined, and based on the valid commodity retrieval data of the test user, the number of commodity retrievals by the test user in multiple historical time periods is determined; for each of the test users, based on the number of commodity purchases, the number of additional purchases and the number of commodity retrievals by the test user in multiple historical time periods, it is determined whether the test user is the test user corresponding to the commodity category.

[0007] Furthermore, the product retrieval data of the test user is screened to determine valid product retrieval data, including: for each product retrieval data, based on the time point corresponding to each product retrieval data of the test user, determining the adjacent product retrieval data of the product retrieval data, determining the retrieval characteristics of the product retrieval data, wherein the retrieval characteristics include the total browsing time, the total number of browsed retrieved products, the total number of clicked retrieved products, and the browsing time of each clicked retrieved product, determining the retrieval keywords of the product retrieval data and the retrieval keywords of the adjacent product retrieval data, calculating the semantic distance between the product retrieval data and the adjacent product retrieval data, and judging whether the product retrieval data is valid product retrieval data according to the retrieval characteristics of the product retrieval data and the semantic distance between the product retrieval data and the adjacent product retrieval data.

[0008] Furthermore, based on the test user's purchase times of commodity categories, the number of times of additional purchases of commodity categories, and the number of times of searches for commodity categories in multiple historical time periods, determining whether the test user is a test user corresponding to the commodity category includes: performing variational mode decomposition on the test user's purchase times of commodity categories in multiple historical time periods to determine multiple purchase feature components, and determining component features corresponding to each of the purchase feature components, wherein the component features at least include a mean, a component standard deviation, a component center frequency, and a component bandwidth; performing variational mode decomposition on the test user's additional purchase times of commodity categories in multiple historical time periods to determine multiple additional purchase feature components, and determining component features corresponding to each of the additional purchase feature components; performing variational mode decomposition on the test user's searches for commodity categories in multiple historical time periods to determine multiple search feature components, and determining component features corresponding to each of the search feature components; based on the component features corresponding to each of the purchase feature components, the component features corresponding to each of the additional purchase feature components, and the component features corresponding to each of the search feature components, determining whether the test user is a test user corresponding to the commodity category.

[0009] Furthermore, based on the component features corresponding to each of the purchase feature components, the component features corresponding to each of the additional purchase feature components, and the component features corresponding to each of the retrieval feature components, determining whether the test user is a test user corresponding to the commodity category includes: generating a purchase component feature matrix based on the component features corresponding to each of the purchase feature components, and calculating purchase feature parameters based on the difference between the purchase component feature matrix and a preset purchase component feature matrix; generating an additional purchase component feature matrix based on the component features corresponding to each of the additional purchase feature components, and calculating additional purchase feature parameters based on the difference between the additional purchase component feature matrix and a preset additional purchase component feature matrix; generating a retrieval component feature matrix based on the component features corresponding to each of the retrieval feature components, and calculating retrieval feature parameters based on the difference between the retrieval component feature matrix and a preset retrieval component feature matrix; performing weighted summation on the purchase feature parameters, additional purchase feature parameters, and retrieval feature parameters to calculate a matching parameter of the test user corresponding to the commodity category, and determining that the test user is a test user corresponding to the commodity category when the matching parameter of the test user corresponding to the commodity category is greater than a matching parameter threshold.

[0010] Furthermore, based on the multiple test users corresponding to each of the commodity categories, the commodity category related parameters of the two commodity categories are calculated, including: S11, taking one of the two commodity categories as the first commodity category, and taking the other commodity category as the second commodity category; S12, sampling a test user from the first commodity category as the first test user, and sampling a test user from the second commodity category as the second test user; S13, based on the component features corresponding to each of the purchase feature components of the first commodity category corresponding to the first test user, the component features corresponding to each of the add-to-purchase feature components, and the component features corresponding to each of the retrieval feature components, and the component features corresponding to each of the purchase feature components of the second test user corresponding to the second commodity category Calculate the user association parameters of the first test user and the second test user corresponding to the first commodity category and the second commodity category based on the component features corresponding to the feature components, the component features corresponding to each of the add-to-cart feature components, and the component features corresponding to each of the retrieval feature components; S14, based on the user association parameters of the first test user and the second test user corresponding to the first commodity category and the second commodity category in each sampling, determine whether to perform sampling again, if so, execute S12, if not, execute S15; S15, based on the user association parameters of the first test user and the second test user corresponding to the first commodity category and the second commodity category in each sampling, calculate the commodity category related parameters of the first commodity category and the second commodity category.

[0011] Furthermore, based on the user portraits of the multiple test users and their shopping behavior data on the e-commerce platform, the multiple test users are clustered to determine a plurality of user clusters, including: calculating the user similarity between any two test users based on their user portraits and their shopping behavior data on the e-commerce platform; and clustering the multiple test users based on the user similarity between any two test users through a clustering algorithm to determine a plurality of user clusters.

[0012] Furthermore, based on the user portraits of the multiple test users and their shopping behavior data on the e-commerce platform, the user similarity between any two test users is calculated, including: based on the user portraits of the multiple test users, calculating the user portrait similarity between any two test users; based on the user association parameters of the multiple test users corresponding to each commodity category, calculating the commodity category association similarity between any two test users; based on the user portrait similarity and commodity category association similarity between any two test users, calculating the user similarity between any two test users.

[0013] Furthermore, based on the shopping behavior data of the user to be analyzed on the e-commerce platform and the multiple user clusters, a user profile of the user to be analyzed is established, including: based on the shopping behavior data of the user to be analyzed on the e-commerce platform, a basic user profile of the user to be analyzed is established; based on the basic user profile of the user to be analyzed, a matching user cluster is determined from the multiple user clusters; based on the user portraits of the multiple test users included in the matching user cluster, the basic user profile of the user to be analyzed is completed to establish a user profile of the user to be analyzed.

[0014] Furthermore, based on the user portrait of the user to be analyzed, the product characteristics of at least one promotional product corresponding to each of the preferential information to be pushed, multiple product categories and product category-related parameters of any two product categories, the pushed products and / or preferential information to be pushed that match the user to be analyzed are determined, including: determining a matching test user from the multiple test users based on the user portrait of the user to be analyzed and the user portraits of multiple test users; determining the first product category of interest to the user to be analyzed based on the user association parameters and multiple product categories corresponding to each product category of the matching test user; determining the second product category of interest to the user to be analyzed based on the first product category of interest to the user to be analyzed and product category-related parameters of any two product categories; determining the pushed products that match the user to be analyzed based on product characteristic information of multiple products on sale on the e-commerce platform, the user portrait of the user to be analyzed, the first product category of interest to the user to be analyzed and the second product category of interest; and / or determining the preferential information to be pushed that matches the user to be analyzed based on the product characteristics of at least one promotional product corresponding to each of the preferential information to be pushed, the user portrait of the user to be analyzed, the first product category of interest to the user to be analyzed and the second product category of interest.

[0015] Compared with the prior art, the method for accurately pushing information to users in a private domain using big data provided in this specification has at least the following beneficial effects: 1. Through in-depth analysis of users' shopping behavior data and user portraits, users' needs, preferences and behavior patterns can be understood more accurately. In this way, the products and discount information pushed to users will be closer to their actual needs, thereby improving user experience and increasing user satisfaction and loyalty. Precision push means showing users products and discounts that they are more likely to be interested in. This personalized recommendation reduces the time users spend filtering information and improves the efficiency and conversion rate of purchase decisions. For e-commerce platforms, this means higher sales and better business performance. Through continuous precision push, e-commerce platforms can maintain close interaction with users and enhance user stickiness. Users will be more willing to stay on the platform to shop because they feel the platform's attention and personalized service to them. Traditional marketing strategies are often cast a wide net, inefficient and costly. The precision push method can accurately target the target user group and achieve precision marketing. This can not only improve marketing efficiency, but also save marketing costs and achieve a higher return on investment. By calculating the relevant parameters between commodity categories, potential associations between different commodity categories can be discovered. This association can help e-commerce platforms consider the diverse needs of users when pushing, promote cross-selling between commodity categories, and increase overall sales. The entire push process is based on big data analysis, which provides strong data support for the decision-making of e-commerce platforms. Whether it is product pricing, promotion strategy or inventory management, these data can be used to make more scientific and reasonable decisions.

[0016] 2. By screening effective product retrieval data, noise data that may be generated by misoperation, invalid search, etc. is excluded, improving the accuracy and reliability of subsequent analysis. Using semantic distance to judge the effectiveness of product retrieval data can more accurately identify the user's true intentions and points of interest. Analyzing users' purchase, add-to-cart and retrieval behaviors in multiple historical time periods can capture the changing trend of user interests over time, thereby making more accurate push notifications. Using methods such as variational mode decomposition to process time series data can extract feature components of different time scales, further enhancing time sensitivity and the depth of analysis. Based on the user's purchase, add-to-cart and retrieval feature components in multiple time periods, combined with their corresponding component features (such as mean, standard deviation, center frequency, etc.), the user's personalized characteristics can be more comprehensively characterized.

[0017] 3. By calculating the user association parameters between two product categories, we can gain a deeper understanding of the potential connections between different product categories. This connection may be based on factors such as users’ common interests, purchasing habits, or similar needs. It helps e-commerce platforms discover the complementary or substitutable relationship between products and provide data support for subsequent marketing strategies such as cross-selling and bundling sales.

[0018] 4. By combining the actual shopping behavior data of the user to be analyzed on the e-commerce platform to establish a basic user portrait, and completing it based on the user portraits of multiple test users in the matching user cluster, the accuracy and comprehensiveness of the user portrait can be significantly improved. This method not only considers the user's direct behavior data, but also incorporates the behavior patterns and characteristics of similar users, making the user portrait closer to the user's real needs and interests. Accurate user portraits provide strong support for personalized push. By deeply understanding the user's shopping habits, preferences and needs, the e-commerce platform can more accurately recommend relevant products and discount information to users, improving the user's shopping experience and satisfaction. By comprehensively considering multiple dimensions such as the user portrait of the user to be analyzed, the user portrait of the test user, the relevant parameters of the product category, and the product characteristics of the discount information to be pushed, the push content and the user are accurately matched. The multi-dimensional matching method can more comprehensively consider the user's interests and needs and improve the pertinence and effectiveness of the push. By determining the first and second categories of interest of the user to be analyzed, and filtering the pushed products and discount information based on these points of interest, the push hit rate and user acceptance can be significantly improved. Users are more likely to be interested in and willing to buy push content that is highly relevant to their interests, thereby improving the conversion rate and sales of the e-commerce platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] This specification will be further described in the form of exemplary embodiments, which will be described in detail by the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same number represents the same structure, wherein: Figure 1 It is a flowchart of a method for accurately pushing data to private domain users in a big data environment shown in one embodiment of the present application; Figure 2 It is a flowchart of calculating commodity category related parameters of two commodity categories shown in one embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to more clearly illustrate the technical solutions of the embodiments of this specification, the following briefly introduces the drawings required for describing the embodiments.

[0021] Figure 1 is a flowchart of a method for accurately pushing data to private domain users in a big data environment in one embodiment of the present application. Figure 1 As shown, a big data private domain precise user push method may include the following steps.

[0022] Step 111, obtaining product feature information of multiple products on sale on the e-commerce platform.

[0023] Specifically, the product feature information of the goods on sale is multi-dimensional, and includes multiple attributes of the goods. These attributes together constitute the characteristics of the goods and determine how the goods are classified. As an example only, the product feature information of the goods on sale may at least include: Product type: This is the most basic classification dimension, such as clothing, electronic products, household items, etc. Brand: the brand of the product, such as Apple, Nike, Huawei, etc. Function or use: the specific function or use of the product, such as mobile phones for communication and entertainment, rice cookers for cooking, etc.; Material: the main material composition of the product, such as cotton clothing, metal mobile phone casing, etc. Specifications and models: Specific specifications of the product, such as size, capacity, and model; Price range: the price range of the product; Applicable people: the type of people the product is suitable for, such as children's toys, adult clothing, etc.

[0024] Step 112: Classify the multiple commodities on sale based on commodity feature information of the multiple commodities on sale to determine multiple commodity categories.

[0025] Specifically, first, it is necessary to clarify the purpose and basis of classification, that is, to determine which product feature information will be used as the main dimension of classification. As an example only, the classification dimensions may at least include product type, brand, function or use, price range, etc. The selected classification algorithm (e.g., decision tree, K-means clustering, support vector machine, etc.) is applied to process the product feature information and assign the products to different categories.

[0026] Step 113, obtaining user portraits of multiple test users and their shopping behavior data on the e-commerce platform.

[0027] Among them, shopping behavior data at least includes order data, shopping cart data and product search data, and user portraits at least include occupational attributes, age attributes, family status attributes and preference attributes.

[0028] As an example only, the user profile of test user 1 is: User ID: U12345 Occupational attributes: Software development engineer Age attribute: 30 years old Family Status: Married, with 1 child Preferred attributes: Shopping preferences: Prefer technology products, children's educational toys and high-quality daily necessities.

[0029] Brand preference: Prefer brands such as Apple, Sony, Lego and Muji.

[0030] Price sensitivity: They have a certain price tolerance for technology products, but pay more attention to cost-effectiveness when it comes to children's products and daily necessities.

[0031] Shopping frequency: 2-3 times a month on average, mostly on weekends or evenings.

[0032] As an example only, the e-commerce platform shopping behavior data of test user 1 includes: 1. Order data: Order ID: OD001 Order time: 2023-04-15 20:30 Order amount: ¥3,500 Purchased goods: 1 laptop computer (¥1,800) 1 set of super racing suit (¥800).

[0033] Order ID: OD002 Order time: 2023-04-22 14:00 Order amount: ¥500 Purchased goods: 2 Disney children's water cups (¥250) 1 pack of baby wipes (¥250).

[0034] 2. Shopping cart data: Shopping cart ID: CT001 Added time: 2023-04-20 10:00 Product List: 1 pair of noise cancelling headphones 1 Smart TV 1 smart bracelet.

[0035] 3. Product search data: Retrieval time: 2023-04-18 19:00 Keywords: "Children's programming robot" Browse search results: Programmable Robot Set (Click for details) Intelligent robot (added to comparison list) Robot (simple browsing) Retrieval time: 2023-04-21 08:30 Keywords: "high quality kitchen utensils" Browse search results: Zwilling knife set (add to cart) Enameled cast iron pan (add to wishlist) Multi-function cooker (see user reviews for details).

[0036] Step 114 , for each commodity category, based on the shopping behavior data of the multiple test users on the e-commerce platform, determine multiple test users corresponding to the commodity category.

[0037] Specifically include: For each test user, the product search data of the test user is screened to determine valid product search data; For each test user, based on the test user's order data, determine the number of times the test user purchased a product in multiple historical time periods; based on the test user's shopping cart data, determine the number of times the test user added a product in multiple historical time periods; based on the test user's valid product search data, determine the number of times the test user searched for a product in multiple historical time periods; For each test user, based on the test user's purchase frequency of the commodity category, the number of additional purchases of the commodity category, and the number of searches of the commodity category in multiple historical time periods, it is determined whether the test user is a test user corresponding to the commodity category.

[0038] Preferably, screening the test user's product search data to determine valid product search data includes: For each product search data, based on the time point corresponding to each product search data of the test user, the adjacent product search data of the product search data is determined, and the search characteristics of the product search data are determined, wherein the search characteristics include the total browsing time, the total number of browsed searched products, the total number of clicked searched products, and the browsing time of each clicked searched product; the search keywords of the product search data and the search keywords of the adjacent product search data are determined, and the semantic distance between the product search data and the adjacent product search data is calculated; and according to the search characteristics of the product search data and the semantic distance between the product search data and the adjacent product search data, it is judged whether the product search data is valid product search data.

[0039] Specifically, the adjacent product retrieval data may be product retrieval data of the same test user whose retrieval time difference with the current product retrieval data is less than a retrieval time difference threshold and which occurs after a time point corresponding to the current product retrieval data.

[0040] The search keywords are directly derived from the text entered by the user in the search box of the e-commerce platform. The core keywords are extracted from the text entered by the user, and the keywords are expanded using synonym libraries, related word lists, or natural language processing tools (such as Word2Vec, BERT, etc.) to cover more relevant query intents. The search keywords of the product search data and the search keywords of the adjacent product search data are converted into word vectors using models such as Word2Vec and GloVe, and the cosine similarity between the two word vectors is calculated. The higher the similarity, the closer the semantic distance.

[0041] The retrieval validity parameter of the product retrieval data can be calculated based on the retrieval characteristics of the product retrieval data and the semantic distance between the product retrieval data and adjacent product retrieval data, and the product retrieval data whose retrieval validity parameter is greater than the retrieval validity parameter threshold is determined to be valid product retrieval data.

[0042] Preferably, the retrieval validity parameter of the commodity retrieval data can be calculated according to the following formula:

[0043] in, The retrieval validity parameter for the jth item retrieval data of the i-th test user, , , , and is the weight, , , , and greater than 0, , The total browsing time corresponding to the j-th product retrieval data of the i-th test user, To preset the total browsing time, The total number of browsed and retrieved products corresponding to the j-th product retrieval data of the i-th test user, The total number of items retrieved for the preset browse, The total number of clicked retrieved products corresponding to the j-th product retrieval data of the i-th test user, To preset the total number of items to be retrieved by clicking, is the browsing time of the kth clicked retrieved product corresponding to the jth product retrieval data of the i-th test user, is the total number of clicked retrieved products corresponding to the j-th product retrieval data of the i-th test user, is the default browsing time average, is the preset semantic distance, The semantic distance between the j-th product retrieval data and the adjacent product retrieval data of the ith test user.

[0044] Preferably, judging whether the test user is a test user corresponding to the commodity category based on the number of commodity category purchases, the number of commodity category additional purchases, and the number of commodity category searches by the test user in multiple historical time periods includes: Performing variational mode decomposition on the number of times the test user purchased a commodity in multiple historical time periods, determining multiple purchase feature components, and determining component features corresponding to each purchase feature component, wherein the component features at least include a mean, a component standard deviation, a component center frequency, and a component bandwidth; Perform variational mode decomposition on the number of times the test user added items to a product category in multiple historical time periods, determine multiple added items feature components, and determine the component features corresponding to each added items feature component; Perform variational mode decomposition on the number of product category searches by the test user in multiple historical time periods, determine multiple search feature components, and determine the component features corresponding to each search feature component; Based on the component feature corresponding to each purchase feature component, the component feature corresponding to each additional purchase feature component, and the component feature corresponding to each search feature component, it is determined whether the test user is a test user corresponding to the product category.

[0045] Specifically, the variational mode decomposition method decomposes the original signal into multiple modal components (Intrinsic Mode Functions, IMFs) through iterative search. Each modal component represents a component of the original signal with different frequencies and bandwidths. For example, different purchase feature components correspond to different purchase behavior patterns, such as seasonal purchases, promotional responses, daily needs, etc. Multiple purchase feature components have different time scales and frequency characteristics, which can reveal the user's purchase behavior patterns in different time periods. For example, a purchase feature component may represent the user's purchase behavior during holidays, while another purchase feature component may represent the user's purchase behavior during promotional activities.

[0046] Preferably, judging whether the test user is a test user corresponding to the commodity category based on the component feature corresponding to each purchase feature component, the component feature corresponding to each add-to-cart feature component, and the component feature corresponding to each search feature component includes: Based on the component features corresponding to each purchase feature component, a purchase component feature matrix is ​​generated, and based on the difference between the purchase component feature matrix and a preset purchase component feature matrix, a purchase feature parameter is calculated, wherein a row vector of the purchase component feature matrix includes a component feature corresponding to a purchase feature component; Generate an additional purchase component feature matrix based on the component feature corresponding to each additional purchase feature component, and calculate the additional purchase feature parameter based on the difference between the additional purchase component feature matrix and the preset additional purchase component feature matrix, wherein a row vector of the additional purchase component feature matrix includes a component feature corresponding to an additional purchase feature component; Based on the component features corresponding to each retrieval feature component, a retrieval component feature matrix is ​​generated, and based on the difference between the retrieval component feature matrix and a preset retrieval component feature matrix, a retrieval feature parameter is calculated, wherein a row vector of the retrieval component feature matrix includes a component feature corresponding to a retrieval feature component; The purchase feature parameters, add-to-cart feature parameters and search feature parameters are weightedly summed to calculate the matching parameter of the test user's corresponding product category. When the matching parameter of the test user's corresponding product category is greater than the matching parameter threshold, the test user is determined to be a test user corresponding to the product category.

[0047] As an example, the purchase characteristic parameter can be calculated according to the following formula:

[0048] in, is the purchase characteristic parameter of the i-th test user, is the preset parameter, greater than 0, is the weight corresponding to the e-th retrieval feature component, is the weight corresponding to the fth component feature, is the value of the element in the e-th row and f-th column of the purchase component feature matrix of the i-th test user, is the value of the element in the e-th row and f-th column of the preset purchase component feature matrix, is the total number of rows included in the purchase component feature matrix, The total number of columns included in the purchase component feature matrix.

[0049] The calculation method of the purchase feature parameters and the search feature parameters is similar to the calculation method of the purchase feature parameters, and will not be repeated here.

[0050] Step 115 : For any two commodity categories, commodity category related parameters of the two commodity categories are calculated based on a plurality of test users corresponding to each commodity category.

[0051] Specifically, the commodity category related parameters of two commodity categories may represent the possibility of a user making a joint purchase, search, or additional purchase of the two commodity categories.

[0052] Figure 2 is a flowchart of calculating commodity category related parameters of two commodity categories shown in an embodiment of the present application, such as Figure 2 As shown, as a preferred embodiment, step 115 specifically includes: S11. One of the two commodity categories is used as the first commodity category, and the other commodity category is used as the second commodity category; S12, sampling a test user from the first commodity category as a first test user, and sampling a test user from the second commodity category as a second test user; S13, calculating user association parameters corresponding to the first commodity category and the second commodity category for the first test user and the second test user based on the component features corresponding to each purchase feature component, each additional purchase feature component, and each search feature component of the first test user and the component features corresponding to each purchase feature component, each additional purchase feature component, and each search feature component of the second test user; S14, based on the user association parameters of the first test user and the second test user corresponding to the first commodity category and the second commodity category in each sampling, determine whether to perform sampling again, if yes, execute S12, if no, execute S15; S15. Calculate commodity category related parameters of the first commodity category and the second commodity category based on the user association parameters of the first test user and the second test user corresponding to the first commodity category and the second commodity category in each sampling.

[0053] Specifically, the user association parameters of the first test user and the second test user corresponding to the first commodity category and the second commodity category can be calculated based on the purchase component feature matrix, added purchase component feature matrix and retrieval component feature matrix of the first test user corresponding to the first commodity category and the purchase component feature matrix, added purchase component feature matrix and retrieval component feature matrix of the second test user corresponding to the second commodity category.

[0054] Preferably, the user association parameters of the first test user and the second test user corresponding to the first commodity category and the second commodity category may be calculated according to the following formula:

[0055] in, The user association parameters for the first and second commodity categories corresponding to the first and second test users are as follows: , and is the weight, , and greater than 0, , , and is the preset parameter, , and greater than 0, is the matrix difference between the purchase component feature matrix of the first test user corresponding to the first commodity category and the purchase component feature matrix of the second test user corresponding to the second commodity category, is the matrix difference between the feature matrix of the additional purchase component corresponding to the first commodity category of the first test user and the feature matrix of the additional purchase component corresponding to the second commodity category of the second test user, It is the matrix difference between the retrieval component feature matrix of the first test user corresponding to the first commodity category and the retrieval component feature matrix of the second test user corresponding to the second commodity category.

[0056] The matrix difference between the purchase component feature matrix of the first test user corresponding to the first commodity category and the purchase component feature matrix of the second test user corresponding to the second commodity category can be calculated according to the following formula:

[0057] in, is the preset parameter, greater than 0, is the value of the element in the e-th row and f-th column of the purchase component feature matrix of the first test user corresponding to the first commodity category, is the value of the element in the e-th row and f-th column of the purchase component feature matrix of the second test user corresponding to the second commodity category.

[0058] The calculation method of the matrix difference between the add-to-cart component feature matrix of the first test user corresponding to the first commodity category and the add-to-cart component feature matrix of the second test user corresponding to the second commodity category, and the matrix difference between the retrieval component feature matrix of the first test user corresponding to the first commodity category and the retrieval component feature matrix of the second test user corresponding to the second commodity category is similar to the calculation method of the matrix difference between the purchase component feature matrix of the first test user corresponding to the first commodity category and the purchase component feature matrix of the second test user corresponding to the second commodity category, and will not be repeated here.

[0059] The user association parameter variance is calculated based on the user association parameters of the first test user and the second test user corresponding to the first commodity category and the second commodity category in each sampling. When the user association parameter variance is less than the user association parameter variance threshold, it is determined not to perform sampling again.

[0060] Based on the user association parameters of the first test user and the second test user corresponding to the first commodity category and the second commodity category in each sampling, the average of the user association parameters is calculated as the commodity category related parameters of the first commodity category and the second commodity category.

[0061] Step 115 , clustering the multiple test users according to their user portraits and shopping behavior data on the e-commerce platform to determine multiple user clusters.

[0062] Specifically include: Calculate the user similarity between any two test users based on the user portraits of multiple test users and their shopping behavior data on the e-commerce platform; A clustering algorithm (eg, K-means clustering, etc.) is used to cluster multiple test users according to user similarities between any two test users to determine multiple user clusters.

[0063] Preferably, the user similarity between any two test users is calculated based on the user portraits of multiple test users and their shopping behavior data on the e-commerce platform, including: Based on the user portraits of multiple test users, calculate the similarity of the user portraits of any two test users; Calculate the commodity category association similarity between any two test users based on the user association parameters of each commodity category for multiple test users; The user similarity between any two test users is calculated based on the user portrait similarity and product category association similarity between any two test users.

[0064] Specifically, the cosine similarity of the user portraits of the two test users may be calculated as the user portrait similarity of the two test users.

[0065] Preferably, the commodity category association similarity of two test users can be calculated according to the following formula:

[0066] in, is the similarity between the product categories of the i-th test user and the j-th test user, is the user association parameter for the i-th test user corresponding to the r-th product category, is the user association parameter for the jth test user corresponding to the rth product category, is the total number of commodity categories, is the preset parameter, Greater than 0.

[0067] The user portrait similarity and commodity category association similarity of the two test users may be weightedly summed to serve as the user similarity of the two test users.

[0068] Step 116, obtaining shopping behavior data of at least one user to be analyzed on the e-commerce platform.

[0069] Step 117: For each user to be analyzed, a user profile of the user to be analyzed is established based on the shopping behavior data of the user to be analyzed on the e-commerce platform and multiple user clusters.

[0070] Specifically include: Based on the shopping behavior data of the users to be analyzed on the e-commerce platform, establish a basic user profile of the users to be analyzed; Based on the basic user profile of the user to be analyzed, a matching user cluster is determined from multiple user clusters; According to the user portraits of multiple test users included in the matching user cluster, the basic user portrait of the user to be analyzed is completed to establish the user portrait of the user to be analyzed.

[0071] Specifically, the cosine similarity between the basic user portrait of the user to be analyzed and the user portrait of the test user corresponding to the cluster center of the user cluster may be calculated, and the user cluster whose cosine similarity is greater than the first cosine similarity threshold may be taken as the matching user cluster.

[0072] The cosine similarity between the basic user portrait of the user to be analyzed and the user portrait of the test user included in the matching user cluster may be calculated, and the test users whose cosine similarity is greater than the second cosine similarity threshold are taken as matching test users.

[0073] The portrait completion model can be used to complete the basic user portrait of the user to be analyzed according to the user portrait of the matching test user, and establish the user portrait of the user to be analyzed. The portrait completion model can be a Generative Adversarial Nets (GAN) model.

[0074] Step 118, obtaining at least one piece of preferential information to be pushed.

[0075] Step 119: for each piece of preferential information to be pushed, analyze the preferential information to be pushed to determine the commodity characteristics of at least one promotional commodity corresponding to the preferential information to be pushed.

[0076] Step 120, based on the user portrait of the user to be analyzed, the product features of at least one promotional product corresponding to each piece of discount information to be pushed, multiple product categories and product category-related parameters of any two product categories, determine the pushed products and / or discount information to be pushed that match the user to be analyzed.

[0077] Specifically include: Based on the user profile of the user to be analyzed and the user profiles of multiple test users, determining a matching test user from the multiple test users; Determine the first commodity class of interest for the user to be analyzed according to the user association parameter of each commodity class corresponding to the matching test user and the multiple commodity classes, for example, taking the commodity class whose user association parameter corresponding to the matching test user is greater than the user association parameter threshold as the first commodity class of interest for the user to be analyzed; Determine the second commodity class of interest to the user to be analyzed based on the first commodity class of interest to the user to be analyzed and commodity class related parameters of any two commodity classes, for example, determine the commodity class whose commodity class related parameters to the first commodity class of interest are greater than a commodity class related parameter threshold as the second commodity class of interest to the user to be analyzed; Determine the pushed products that match the user to be analyzed based on product feature information of multiple products on sale on the e-commerce platform, the user portrait of the user to be analyzed, the first product category and the second product category of interest to the user to be analyzed; and / or, Based on the product features of at least one promotional product corresponding to each piece of discount information to be pushed, the user portrait of the user to be analyzed, the first product category and the second product category of interest to the user to be analyzed, the discount information to be pushed that matches the user to be analyzed is determined.

[0078] Specifically, the product matching model can be used to determine the pushed products that match the user to be analyzed based on the product feature information of multiple products on sale on the e-commerce platform, the user portrait of the user to be analyzed, the first product category of interest, and the second product category of interest of the user to be analyzed. The product matching model can be a convolutional neural network model.

[0079] The discount matching model can be used to determine the discount information to be pushed that matches the user to be analyzed based on the product characteristics of at least one promotional product corresponding to each discount information to be pushed, the user portrait of the user to be analyzed, the first product category of interest, and the second product category of interest of the user to be analyzed, wherein the discount matching model can be a convolutional neural network model.

[0080] Finally, it should be understood that the embodiments described in this specification are only used to illustrate the principles of the embodiments of this specification. Other variations may also fall within the scope of this specification. Therefore, as an example and not a limitation, alternative configurations of the embodiments of this specification may be considered consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to the embodiments explicitly introduced and described in this specification.

Claims

1. A method for accurately pushing data to private domain users based on big data, characterized in that: include: Obtain product feature information of multiple products on sale on the e-commerce platform; Based on the commodity feature information of the plurality of commodities on sale, the plurality of commodities on sale are classified to determine a plurality of commodity categories; Obtaining user portraits of multiple test users and shopping behavior data on the e-commerce platform, wherein the shopping behavior data at least includes order data, shopping cart data, and product search data, and the user portraits at least include occupation attributes, age attributes, family status attributes, and preference attributes; For each of the commodity categories, determining a plurality of test users corresponding to the commodity category based on the shopping behavior data of the plurality of test users on the e-commerce platform; For any two of the commodity categories, based on a plurality of test users corresponding to each of the commodity categories, commodity category related parameters of the two commodity categories are calculated; Clustering the multiple test users according to their user portraits and their shopping behavior data on the e-commerce platform to determine multiple user clusters; Obtaining shopping behavior data of at least one user to be analyzed on the e-commerce platform; For each of the to-be-analyzed users, establishing a user profile of the to-be-analyzed user based on the shopping behavior data of the to-be-analyzed user on the e-commerce platform and the multiple user clusters; Get at least one piece of discount information to be pushed; For each piece of the preferential information to be pushed, analyzing the preferential information to be pushed, and determining the commodity characteristics of at least one promotional commodity corresponding to the preferential information to be pushed; Based on the user portrait of the user to be analyzed, the product characteristics of at least one promotional product corresponding to each piece of the discount information to be pushed, multiple product categories and product category-related parameters of any two product categories, determine the pushed products and / or discount information to be pushed that match the user to be analyzed.

2. According to claim 1, a method for accurately pushing data to private domains using big data is characterized in that: Determining a plurality of test users corresponding to the commodity category based on the shopping behavior data of the plurality of test users on the e-commerce platform includes: For each of the test users, screening the commodity search data of the test user to determine valid commodity search data; For each of the test users, based on the order data of the test user, determine the number of times the test user purchased a product in multiple historical time periods; based on the shopping cart data of the test user, determine the number of times the test user added a product in multiple historical time periods; based on the valid product search data of the test user, determine the number of times the test user searched for a product in multiple historical time periods; For each of the test users, based on the test user's purchase frequency of the commodity category, the number of additional purchases of the commodity category, and the number of searches for the commodity category in multiple historical time periods, it is determined whether the test user is a test user corresponding to the commodity category.

3. A method for accurately pushing data to private domain users based on big data according to claim 2, characterized in that: Screening the test user's product search data to determine valid product search data includes: For each of the product retrieval data, based on the time point corresponding to each of the product retrieval data of the test user, determine the adjacent product retrieval data of the product retrieval data, determine the retrieval features of the product retrieval data, wherein the retrieval features include the total browsing time, the total number of browsed and retrieved products, the total number of clicked retrieved products, and the browsing time of each clicked retrieved product, determine the retrieval keywords of the product retrieval data and the retrieval keywords of the adjacent product retrieval data, calculate the semantic distance between the product retrieval data and the adjacent product retrieval data, and judge whether the product retrieval data is valid product retrieval data according to the retrieval features of the product retrieval data and the semantic distance between the product retrieval data and the adjacent product retrieval data.

4. According to claim 2, a method for accurately pushing data to private domains using big data is characterized in that: Based on the number of commodity category purchases, the number of commodity category additional purchases, and the number of commodity category searches by the test user in multiple historical time periods, determining whether the test user is a test user corresponding to the commodity category includes: Performing variational mode decomposition on the number of times the test user purchased a commodity in multiple historical time periods, determining multiple purchase feature components, and determining component features corresponding to each of the purchase feature components, wherein the component features at least include a mean, a component standard deviation, a component center frequency, and a component bandwidth; Performing variational mode decomposition on the number of times the test user added items to a product category in multiple historical time periods to determine multiple added items feature components, and determining component features corresponding to each of the added items feature components; Performing variational mode decomposition on the number of commodity category searches by the test user in multiple historical time periods to determine multiple search feature components, and determining component features corresponding to each of the search feature components; Based on the component feature corresponding to each of the purchase feature components, the component feature corresponding to each of the additional purchase feature components, and the component feature corresponding to each of the search feature components, it is determined whether the test user is a test user corresponding to the commodity category.

5. A method for accurately pushing data to private domain users based on big data according to claim 4, characterized in that: Based on the component feature corresponding to each of the purchase feature components, the component feature corresponding to each of the additional purchase feature components, and the component feature corresponding to each of the search feature components, determining whether the test user is a test user corresponding to the commodity category includes: Generate a purchase component feature matrix based on the component features corresponding to each of the purchase feature components, and calculate purchase feature parameters based on the difference between the purchase component feature matrix and a preset purchase component feature matrix; Generate an additional purchase component feature matrix based on the component feature corresponding to each of the additional purchase feature components, and calculate additional purchase feature parameters based on the difference between the additional purchase component feature matrix and a preset additional purchase component feature matrix; Generate a retrieval component feature matrix based on the component features corresponding to each of the retrieval feature components, and calculate retrieval feature parameters based on the difference between the retrieval component feature matrix and a preset retrieval component feature matrix; A weighted sum is performed on the purchase feature parameters, the additional purchase feature parameters and the search feature parameters to calculate the matching parameter of the test user corresponding to the commodity category; when the matching parameter of the test user corresponding to the commodity category is greater than a matching parameter threshold, the test user is determined to be a test user corresponding to the commodity category.

6. A method for accurately pushing data to private domain users based on big data according to claim 4, characterized in that: Calculating commodity category related parameters of two commodity categories based on a plurality of test users corresponding to each commodity category, including: S11, one of the two commodity categories is used as the first commodity category, and the other commodity category is used as the second commodity category; S12: sampling a test user from the first commodity category as a first test user, and sampling a test user from the second commodity category as a second test user; S13, calculating user association parameters of the first test user and the second test user corresponding to the first commodity category and the second commodity category based on the component features corresponding to each of the purchase feature components, the component features corresponding to each of the additional purchase feature components, and the component features corresponding to each of the retrieval feature components of the first test user corresponding to the first commodity category and the component features corresponding to each of the purchase feature components, the component features corresponding to each of the additional purchase feature components, and the component features corresponding to each of the retrieval feature components of the second test user corresponding to the second commodity category; S14, based on the user association parameters of the first test user and the second test user corresponding to the first commodity category and the second commodity category in each sampling, determine whether to perform sampling again, if yes, execute S12, if no, execute S15; S15. Calculate commodity category related parameters of the first commodity category and the second commodity category based on the user association parameters of the first test user and the second test user sampled in each time corresponding to the first commodity category and the second commodity category.

7. A method for accurately pushing data to private domain users based on big data according to claim 6, characterized in that: Clustering the multiple test users according to their user portraits and their shopping behavior data on the e-commerce platform to determine multiple user clusters includes: Calculate the user similarity between any two test users based on the user portraits of the multiple test users and their shopping behavior data on the e-commerce platform; The plurality of test users are clustered according to the user similarity between any two test users by using a clustering algorithm to determine a plurality of user clusters.

8. A method for accurately pushing data to private domain users based on big data according to claim 7, characterized in that: Calculating the user similarity between any two test users based on the user portraits of the multiple test users and their shopping behavior data on the e-commerce platform includes: Calculate the similarity of user portraits of any two test users based on the user portraits of the multiple test users; Calculating the commodity category association similarity between any two test users according to the user association parameters of the multiple test users corresponding to each commodity category; The user similarity between any two test users is calculated based on the user portrait similarity and product category association similarity between any two test users.

9. A method for accurately pushing private domain data to users according to any one of claims 1 to 8, characterized in that: Based on the shopping behavior data of the user to be analyzed on the e-commerce platform and the multiple user clusters, a user profile of the user to be analyzed is established, including: Based on the shopping behavior data of the user to be analyzed on the e-commerce platform, a basic user profile of the user to be analyzed is established; Based on the basic user profile of the user to be analyzed, determining a matching user cluster from the multiple user clusters; According to the user portraits of the multiple test users included in the matching user cluster, the basic user portrait of the user to be analyzed is completed to establish the user portrait of the user to be analyzed.

10. A method for accurately pushing data to private domain users based on big data according to claim 8, characterized in that: Determining the pushed products and / or the to-be-pushed preferential information matching the to-be-analyzed user based on the user portrait of the to-be-analyzed user, the product features of at least one promotional product corresponding to each of the to-be-pushed preferential information, multiple product categories, and product category-related parameters of any two product categories, including: Based on the user portrait of the user to be analyzed and the user portraits of multiple test users, determining a matching test user from the multiple test users; Determining a first commodity category of interest to the user to be analyzed according to the user association parameter corresponding to each commodity category of the matching test user and the multiple commodity categories; Determining a second commodity category of interest to the user to be analyzed based on the first commodity category of interest to the user to be analyzed and commodity category related parameters of any two commodity categories; Determine the pushed products that match the user to be analyzed based on product feature information of multiple products on sale on the e-commerce platform, the user portrait of the user to be analyzed, the first product category and the second product category of interest to the user to be analyzed; and / or, Based on the product features of at least one promotional product corresponding to each piece of the discount information to be pushed, the user portrait of the user to be analyzed, the first product category and the second product category of interest of the user to be analyzed, the discount information to be pushed that matches the user to be analyzed is determined.

Images