Intelligent matching commodity collaborative filtering recommendation method based on user portrait
Through the intelligent order allocation method based on user portrait, the correlation between products and portrait correlation is calculated, and the existing order allocation efficiency and high transportation cost are solved, achieving more efficient order allocation and better user experience.
Patent Information
- Application Number
- CN202510614899.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-14
- Publication Date
- 2025-06-13
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing intelligent order distribution technology is inefficient, has high transportation costs and poor user experience when handling user orders, especially when shipped in multiple warehouses.
The intelligent order distribution product collaborative filtering recommendation method based on user portraits is adopted. By obtaining the product listing time and user order data, the purchase correlation and portrait correlation between products are calculated, and combined with the portrait correlation credibility index, the product distribution is optimized.
It improves order distribution efficiency, reduces transportation costs, improves user experience, and reduces packaging and sorting costs by rationally selecting the shipping warehouse.
Smart Images
Figure CN120146969A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data processing, and particularly to an intelligent order matching commodity collaborative filtering recommendation method based on user portraits. Background Art
[0002] Intelligent order matching is an automated order allocation technology based on artificial intelligence and big data analysis, which is widely used in industries such as e-commerce, logistics, and catering. By analyzing various order matching factors, it intelligently matches the optimal suppliers, warehouses, or delivery routes, etc., thereby improving order processing efficiency, reducing costs, and optimizing the user experience.
[0003] In the current order matching process, since there are often highly relevant commodities in a single user order, and highly relevant commodities often have similar demand scenarios or uses. If shipped from multiple warehouses respectively, it will not only increase transportation costs, but also may lead to asynchronous delivery times, affecting the overall user experience. In addition, shipping from multiple warehouses will also increase packaging and sorting costs and reduce order fulfillment efficiency. Summary of the Invention
[0004] The present invention provides an intelligent order matching commodity collaborative filtering recommendation method based on user portraits to solve the problems of low existing order matching efficiency, high transportation costs, and poor user experience. The specific technical solutions adopted are as follows: The present invention proposes an intelligent order matching commodity collaborative filtering recommendation method based on user portraits, and the method includes the following steps: Obtain the shelf time of the commodity; obtain the commodities, order times, and user portrait tags of all orders of each user; obtain the purchase correlation degree between every two commodities according to the number of orders in which the two commodities are purchased simultaneously. Screen out newly listed commodities according to the shelf time of the commodity and the purchase correlation degree, and denote any one of the newly listed commodities as the target commodity; obtain the portrait correlation degree between the target commodity and each other commodity in the orders of users with each portrait tag according to the situation of simultaneous purchase. Obtain the association confidence degree of each portrait tag for the target commodity and each other commodity according to the difference situation of the portrait correlation degree between the target commodity and each other commodity in each portrait tag; obtain the enthusiasm degree of each portrait tag for purchasing the target commodity according to the interval situation between the order time of the user purchasing the target commodity with each portrait tag and the shelf time of the target commodity; combine the association confidence degree to obtain the association credibility index of each portrait tag for the target commodity and each other commodity; obtain the association index of the target commodity and each other commodity according to the portrait correlation degree between the target commodity and each other commodity in each portrait tag, and the association credibility index of each portrait tag for the target commodity and each other commodity. Match the orders of the user according to the purchase correlation and the correlation index.
[0005] Further, the method for screening new products according to the listing time of the product and the purchase correlation includes the following specific steps: Mark the products that have been listed within the last month and have a purchase correlation with all other products less than the preset correlation threshold as new products.
[0006] Further, the method for obtaining the portrait correlation degree between the target product and each other product in each portrait label according to the situation where the target product and other products are purchased simultaneously in the orders of users with each portrait label includes the following specific steps: In the formula, is the portrait correlation degree between the target product and the th product other than the target product in the th portrait label; is the number of orders in which the target product and the th product other than the target product are purchased simultaneously among all orders of all users with the th portrait label; is the number of people who purchase the target product and the th product other than the target product simultaneously among all users with the th portrait label; is the number of users with the th portrait label.
[0007] Further, the method for obtaining the correlation confidence degree of each portrait label for the target product and each other product according to the difference situation of the portrait correlation degree between the target product and other products in each portrait label includes the following specific steps: In the formula, is the correlation confidence degree of the th portrait label for the target product and the th product other than the target product; is the portrait correlation degree between the target product and the th product other than the target product in the th portrait label; is the average value of the portrait correlation degrees between the target product and the th product other than the target product in all portrait labels; is the linear normalization function.
[0008] Further, obtaining the enthusiasm degree of each portrait label for purchasing the target product according to the interval between the order time of the user purchasing the target product corresponding to each portrait label and the listing time of the target product, the specific method included is as follows: For any user who has purchased the target product, the difference obtained by subtracting the listing time of the target product from the order time of the user purchasing the target product is denoted as the enthusiasm index of the user for purchasing the target product; For any portrait label, the average value of the enthusiasm indexes of all users of the portrait label for purchasing the target product is denoted as the enthusiasm degree of the portrait label for purchasing the target product.
[0009] Further, obtaining the correlation credibility index of each portrait label for the target product and each other product, the specific method included is as follows: For any portrait label and any product other than the target product, the product of the correlation confidence of the portrait label for the target product and the product and the enthusiasm degree of the portrait label for purchasing the target product is denoted as the correlation credibility index of the portrait label for the target product and the product.
[0010] Further, obtaining the correlation index of the target product and each other product according to the portrait correlation degree of the target product and each other product in each portrait label, and the correlation credibility index of each portrait label for the target product and each other product, the specific method included is as follows: In the formula, is the correlation index of the target product and the th product other than the target product; is the th portrait label's correlation credibility index for the target product and the th product other than the target product; is the portrait correlation degree of the target product and the th product other than the target product in the th portrait label; is the total number of portrait labels; is the weight normalization function; is the linear normalization function.
[0011] Further, performing order matching for the user's order according to the purchase correlation degree and the correlation index, the specific method included is as follows: For any user's order, respectively obtain the purchase correlation degree between any two products in the order. If there are newly listed products in the order, obtain the correlation index of each newly listed product and other products; Two products with a purchase correlation greater than a preset correlation threshold are recorded as a pair of related products, and two products with a correlation index greater than a preset correlation threshold are recorded as a pair of related products; Obtain the user's delivery address. For any pair of related products, obtain the warehouse containing the pair of related products in the inventory and the address of the warehouse, and use the warehouse closest to the user's delivery address as the shipping warehouse to ship to the user.
[0012] Further, the method for obtaining the purchase correlation between every two products according to the number of orders in which the two products are purchased simultaneously includes the following specific method: In the formula, is the purchase correlation between the th product and the th product; is the number of orders in which the th product and the th product are purchased simultaneously among all orders of all users; is the number of orders in which the th product is purchased among all orders of all users; is the number of orders in which the th product is purchased among all orders of all users.
[0013] Further, the specific method for obtaining the user's portrait label is as follows: In the database of the e-commerce platform, perform manual category division on all products of the e-commerce platform; Obtain all order data of each user in the past year. The order data includes the product prices of all products purchased in each order; For any user and any category, obtain the linearly normalized result of the total consumption amount of the user in the category in the past year, and record it as the consumption index of the user in the category; the category with the largest consumption index of the user is recorded as the portrait label of the user.
[0014] The beneficial effects of the present invention are as follows: when matching orders for goods through a collaborative filtering algorithm, due to the small amount of sales data of new goods, the purchase correlation between new goods and other goods is too low. The present invention distinguishes users through portrait tags and judges the portrait correlation between new goods and other goods in each portrait tag; since users with different portrait tags have different interests in different categories, different portrait tags have different weights when judging the correlation between new goods and other goods. The present invention obtains the correlation confidence of each portrait tag to the target goods and each other goods through the difference of the portrait correlation between the target goods and other goods in each portrait tag, and then obtains the degree of enthusiasm of each portrait tag to purchase the target goods through the interval between the order time of the user of each portrait tag to purchase the target goods and the shelf time of the target goods, and then obtains the correlation credibility index, and weighs the weights of different portrait tags when judging the correlation between new goods and other goods. So far, the correlation index between the target goods and each other goods is obtained through the portrait correlation and the correlation credibility index, and the goods are matched to the user's order in combination with the purchase correlation, which improves the efficiency of matching, reduces transportation costs and improves user experience. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0016] Figure 1 A flowchart of a collaborative filtering recommendation method for intelligent order matching products based on user portraits provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0018] See also Figure 1 , which shows a flow chart of a method for collaborative filtering and recommending intelligent order matching products based on user portraits provided by an embodiment of the present invention, the method comprising the following steps: Step S001: Obtain the listing time of the product; obtain the products, order times, and user portrait tags of all orders of each user; obtain the purchase correlation degree between every two products according to the number of orders in which the two products are purchased simultaneously.
[0019] It should be noted that collaborative filtering is a recommendation algorithm that predicts user preferences by mining group behavior patterns. The core idea of collaborative filtering is that similar users or items will show similar behavior preferences. Even if the attributes of products are quite different, the correlation between products can be captured through the similarity of product preferences shown in the purchase orders of similar users. In this embodiment, the correlation between products is measured based on collaborative filtering of user portraits. Therefore, it is first necessary to obtain the order data of users' purchased products.
[0020] Specifically, in the database of the e-commerce platform, obtain the listing time of each product, and conduct manual category classification for all products on the e-commerce platform. The categories include but are not limited to "Food and Beverage", "Clothing, Shoes and Hats", "Home Life", "3C Digital", "Beauty and Personal Care", "Maternal and Infant Supplies", "Books and Entertainment", "Automobile Supplies", "Sports and Outdoor". Obtain all order data of each user in the past year. The order data includes the product prices and order times of all products purchased in each order. For any user and any category, obtain the linear normalization result of the total consumption amount of the user in this category in the past year, which is recorded as the consumption index of the user in this category. The normalization object is the total consumption amount of all users in this category in the past year; the category with the largest consumption index of the user is recorded as the portrait tag of the user.
[0021] It should be noted that when selecting the delivery warehouse, mainly products with a higher correlation degree are shipped from the same warehouse to improve the order picking efficiency. Therefore, it is necessary to calculate the correlation degree between every two products. For any two products, if the two products are purchased simultaneously by a large number of users, it indicates that there is a high correlation degree between these two products. Accordingly, calculate the purchase correlation degree between every two products.
[0022] Specifically, the th product and the th product are calculated for the purchase correlation degree as follows: In the formula, is the purchase correlation degree between the th product and the th product; is the number of orders in which the th product and the The number of orders for a product purchased at the same time; For all orders of all users The number of orders for each item purchased; For all orders of all users The number of orders for the product purchased; it should be noted that if the Products and If one of the products has not been purchased in the past year, Products and The purchase relevance of a product is recorded as 0.
[0023] Step S002, filter out new products according to the product listing time and the purchase association, and record any new product as a target product; based on the fact that the target product and other products are purchased at the same time in the orders of users of each portrait tag, obtain the portrait association degree of the target product and each other product in each portrait tag.
[0024] It should be noted that new products will be continuously launched during the continuous operation of the e-commerce platform, and due to the lack of sales data for new products, the purchase correlation between new products and other products is too low. The portrait tags of the e-commerce platform reflect the user's interest preference in the tag field. When a new product in any tag field is launched, ordinary users may not pay attention to and purchase it immediately, while users with portrait tags in the tag field will pay more attention to the new product and are more likely to participate in the purchase. Therefore, based on the similar purchase behavior of users with the same portrait tags, the portrait correlation between the new products of each category and other products is determined.
[0025] Specifically, a product that has been put on the shelf within the past month and whose purchase correlation with all other products is less than a preset correlation threshold is recorded as a new product; wherein the preset correlation threshold is 0.1, and this embodiment is described by taking this as an example; Record any new product as the target product; Target product and other products Products in The calculation method of the portrait relevance of portrait tags is: In the formula, The target product and the Products in The portrait relevance of the portrait tags; For the Among all orders of all users with the same portrait tag, the target product and the second product other than the target product are purchased at the same time. The order quantity of each item; Among all users with the th image tag, the number of people who purchase the target product and the th product other than the target product at the same time; is the number of users with the th image tag.
[0026] It should be noted that the larger the , the larger the number of orders for purchasing the target product and the th product other than the target product at the same time, and the greater the correlation between the target product and the th product other than the target product; the larger the , it indicates that among the users with the th image tag, most users purchase the target product and the th product other than the target product at the same time, and the greater the correlation between the target product and the th product other than the target product. The larger the , the greater the correlation between the target product and the th product other than the target product.
[0027] Step S003: According to the difference in the image correlation degree of the target product and other products in each image tag, obtain the correlation confidence degree of each image tag for the target product and each other product; according to the interval between the order time of the target product purchased by the users of each image tag and the shelf time of the target product, obtain the enthusiasm degree of each image tag for purchasing the target product; combine the said correlation confidence degree to obtain the correlation credibility index of each image tag for the target product and each other product; according to the image correlation degree of the target product and each other product in each image tag, and the correlation credibility index of each image tag for the target product and each other product, obtain the correlation index of the target product and each other product.
[0028] It should be noted that after obtaining the image correlation degree of the target product and other products in each image tag, since the interest degrees of the users of different image tags in the category where the target product is located are different, the image correlation degree of the target product and other products in each image tag has limitations when expressing the correlation degree between the target product and other products. Therefore, when judging the correlation degree between the new product and other products, it is necessary to first judge the correlation confidence degree of each image tag for the target product and other products.
[0029] It should be further noted that for any one image tag and any one product other than the target product, if the difference between the image correlation degree of the target product and this product in this image tag and the image correlation degree in other image tags is greater, it indicates that the image correlation degree of the target product and this product in this image tag can less represent the correlation degree between the target product and this product, and the correlation confidence degree of this image tag for the target product and this product is lower.
[0030] Specifically, the The calculation method of the association confidence degree of a pair of an image label for the target product and the th product other than the target product is as follows: In the formula, is the association confidence degree of a pair of the th image label for the target product and the th product other than the target product; is the image association degree of the target product and the th product other than the target product in the th image label; is the average value of the image association degrees of the target product and the th product other than the target product in all image labels; is a linear normalization function, and the normalization object is the of a pair of the target product and the th product other than the target product for all image labels.
[0031] It should be noted that for users with a relatively high association confidence degree of an image label, if the user makes a purchase immediately after a new product is launched, it indicates that the users of this image label have a relatively high degree of attention to the new product, and these users can better reflect the association information between the new product and other products.
[0032] Specifically, for any user who has purchased the target product, the difference obtained by subtracting the launch time of the target product from the order time of the user purchasing the target product is recorded as the positive index of the user purchasing the target product; For any image label, the average value of the positive indexes of all users of this image label purchasing the target product is recorded as the positive degree of this image label purchasing the target product.
[0033] It should be noted that for any image label and any product other than the target product, if the association confidence degree of this image label for the target product and this product is higher, and the positive degree of this image label purchasing the target product is higher, it indicates that the credibility of the users of this image label when measuring the association degree between the target product and this product is higher. Therefore, the association credibility index of each image label for the target product and other products is calculated accordingly.
[0034] Specifically, for any image label and any product other than the target product, the product of the association confidence degree of this image label for the target product and this product and the positive degree of this image label purchasing the target product is recorded as the association credibility index of this image label for the target product and this product.
[0035] It should be noted that portrait tags with a relatively high correlation trust index can measure the correlation between products more accurately. Therefore, it is necessary to combine the correlation trust index to accurately judge the correlation index between the target product and other products.
[0036] Specifically, the calculation method of the correlation index between the target product and the th product other than the target product is as follows: In the formula, is the correlation index between the target product and the th product other than the target product; is the correlation trust index of the th portrait tag for the target product and the th product other than the target product; is the portrait correlation degree of the target product and the th product other than the target product in the th portrait tag; is the total number of portrait tags; is the weight normalization function, and the normalization object is the correlation trust index of all portrait tags for the target product and the th product other than the target product; is the linear normalization function, and the normalization object is the correlation index between the target product and all products other than the target product.
[0037] Step S004, perform order matching for the user's order according to the purchase correlation degree and the correlation index.
[0038] It should be noted that during the order matching process, since there are often products with high correlation in a single order of the user, products with high correlation often have similar demand scenarios or uses. Shipping products with high correlation from the same warehouse can reduce transportation costs, reduce the problem of asynchronous delivery times, optimize the overall user experience, and at the same time reduce packaging and sorting costs and improve order fulfillment efficiency. Therefore, after obtaining the purchase correlation degree between products and the correlation index between the new product and other products, it is necessary to screen products with relatively high correlation.
[0039] Specifically, for any user's order, obtain the purchase correlation degree between any two products in the order respectively. If there are new products in the order, obtain the correlation index between each new product and other products; Mark two products with a purchase correlation degree greater than the preset correlation threshold as a pair of related products, and mark two products with a correlation index greater than the preset correlation threshold as a pair of related products. Among them, the preset correlation threshold is 0.7, and this embodiment is described by taking this as an example. Obtain the user's delivery address. For any pair of related products, obtain the warehouse that has this pair of related products in stock and the address of the warehouse, and use the warehouse closest to the user's delivery address as the shipping warehouse to ship to the user.
[0040] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principles of the present invention shall be included within the protection scope of the present invention.
Claims
1. The intelligent collaborative filtering recommendation method for order matching products based on user portraits is characterized by: The method comprises the following steps: Get the listing time of the product; get the products, order time and user portrait tags of all orders of each user; get the purchase correlation between every two products based on the number of orders in which every two products are purchased at the same time; New products are obtained by screening according to the product listing time and the purchase association, and any new product is recorded as a target product; according to the fact that the target product and other products are purchased at the same time in the orders of users of each portrait tag, the portrait association degree of the target product and each other product in each portrait tag is obtained; According to the difference in the image association degree between the target product and other products in each image label, the association confidence degree between each image label and the target product and each other product is obtained; according to the interval between the order time of the user of each image label to purchase the target product and the shelf time of the target product, the activeness of each image label to purchase the target product is obtained; combined with the association confidence degree, the association credibility index between each image label and the target product and each other product is obtained; according to the image association degree between the target product and each other product in each image label, and the association credibility index between each image label and the target product and each other product, the association index between the target product and each other product is obtained; The user's order is matched according to the purchase correlation degree and the correlation index.
2. The method for intelligent collaborative filtering recommendation of goods for order matching based on user portrait according to claim 1 is characterized in that: The specific method of selecting and obtaining new products according to the product listing time and the purchase relevance is as follows: Products that have been launched in the past month and whose purchase correlation with all other products is less than the preset correlation threshold will be recorded as new products.
3. The intelligent collaborative filtering recommendation method for order matching products based on user portrait according to claim 1 is characterized in that: The specific method of obtaining the image correlation degree of the target product and each other product in each image tag according to the situation that the target product and other products are purchased at the same time in the order of the user of each image tag is as follows: In the formula, The target product and the Products in The portrait relevance of the portrait tags; For the Among all orders of all users with the same portrait tag, the target product and the second product other than the target product are purchased at the same time. The order quantity of each item; For the Among all users with portrait tags, those who purchased both the target product and the third product other than the target product Number of people who buy a product; For the The number of users with a portrait tag.
4. The method for intelligent collaborative filtering and recommendation of goods for order matching based on user portrait according to claim 1 is characterized in that: The specific method of obtaining the association confidence of each image label with the target product and each other product according to the difference in the image association degree of each image label of the target product and other products includes: In the formula, For the The target product and the first product other than the target product The confidence level of association of each product; The target product and the Products in The portrait relevance of the portrait tags; The target product and the The average value of the image relevance of a product in all image tags; is a linear normalization function.
5. The method for intelligent collaborative filtering and recommendation of goods for order matching based on user portrait according to claim 1 is characterized in that: The method of obtaining the activeness of each portrait label in purchasing the target product according to the order time of the user purchasing the target product and the interval between the time when the target product is put on the shelves by the user of each portrait label includes the following specific methods: For any user who has purchased the target product, the difference between the order time of the user purchasing the target product and the shelf time of the target product is recorded as the user's positive index for purchasing the target product; For any portrait tag, the average of the positive indexes of all users with the portrait tag in purchasing the target product is recorded as the degree of positiveness of purchasing the target product with the portrait tag.
6. The method for intelligent collaborative filtering and recommendation of goods for order matching based on user portrait according to claim 1 is characterized in that: The specific method of obtaining the association trust index of each image label with the target product and each other product includes: For any image tag and any product except the target product, the product's association confidence of the image tag with the target product is multiplied by the image tag's enthusiasm for purchasing the target product, which is recorded as the image tag's association credibility index with the target product.
7. The method for intelligent collaborative filtering and recommendation of goods for order matching based on user portrait according to claim 1 is characterized in that: The method of obtaining the association index between the target product and each other product according to the image association degree between the target product and each other product in each image label and the association credibility index between each image label and the target product and each other product includes: In the formula, The target product and the The correlation index of each product; For the The image label is used to compare the target product with the third The correlation credibility index of each product; The target product and the Products in The portrait relevance of the portrait tags; is the total number of portrait tags; is the weight normalization function; is a linear normalization function.
8. The method for intelligent collaborative filtering and recommendation of goods for order matching based on user portrait according to claim 1 is characterized in that: The specific method of matching the user's order according to the purchase correlation degree and the correlation index is as follows: For any user's order, obtain the purchase correlation between any two products in the order. If there are new products in the order, obtain the correlation index between each new product and other products. Two commodities with a purchase correlation greater than a preset correlation threshold are recorded as a pair of related commodities, and two commodities with a correlation index greater than a preset correlation threshold are recorded as a pair of related commodities; Get the user's delivery address. For any pair of related products, get the warehouse in stock that contains the pair of related products and the address of the warehouse. Use the warehouse closest to the user's delivery address as the shipping warehouse to ship to the user.
9. The method for intelligent collaborative filtering and recommendation of goods for order matching based on user portrait according to claim 1 is characterized in that: The specific method of obtaining the purchase correlation between each two commodities according to the number of orders in which each two commodities are purchased simultaneously includes: In the formula, For the Products and The purchase relevance of each product; For all orders of all users Products and The number of orders for a product purchased at the same time; For all orders of all users The number of orders for each item purchased; For all orders of all users The number of orders for a product that was purchased.
10. The method for intelligent collaborative filtering and recommendation of goods for order matching based on user portrait according to claim 1, characterized in that: The user's portrait tag is specifically obtained by: In the database of the e-commerce platform, all the products on the e-commerce platform are manually divided into categories; Obtain all order data of each user in the past year, wherein the order data includes the commodity prices of all commodities purchased in each order; For any user and any category, obtain the linear normalized result of the user's total consumption in the category in the past year, and record it as the user's consumption index in the category; record the category with the largest consumption index of the user as the user's portrait label.