Item recommendation method and system based on collaborative filtering model

Abstract

The application discloses a commodity recommendation method and system based on a collaborative filtering model, comprising a data collection module for collecting data of users and commodities in an e-commerce platform scene; a data preprocessing module for preprocessing the collected data; a collaborative model training module for training recommendation results based on a user behavior collaborative filtering model; a model verification module for verifying whether the recommendation results conform to the user behavior trend; and a deep formulation module for secondary training to formulate deep personalized commodity specification recommendations, using data mining and machine learning technology to analyze the data, thereby extracting user features, commodity features and scene features, establishing a user-specific interest preference model, and finally finding the commodities of interest to the user from the massive commodities to meet the personalized needs of the user. The application has the characteristics of strong practicability and high personalization.

Description

Technical Field

[0001] This invention relates to the field of e-commerce, specifically to a product recommendation method and system based on a collaborative filtering model. Background Technology

[0002] Currently, online shopping malls bring great convenience to people's lives, allowing them to purchase the goods they need without leaving home. However, with the improvement of living standards, people's needs are gradually developing towards diversification, enjoyment, and personalization. Faced with numerous products, when users' needs are not clear, it is difficult for them to find suitable products. Personalized e-commerce recommendations can effectively solve this problem, proactively providing users with products they might like, like a shopping guide, thereby improving shopping efficiency, meeting personalized needs, and enhancing user experience. However, existing product recommendation systems lack a high degree of personalization in their recommendations, failing to intuitively leave users with a purchase intention, resulting in low conversion rates. Therefore, it is essential to design a practical and highly personalized product recommendation method and system based on a collaborative filtering model. Summary of the Invention

[0003] The purpose of this invention is to provide a product recommendation method and system based on a collaborative filtering model to solve the problems mentioned in the background art.

[0004] To address the aforementioned technical problems, this invention provides the following technical solution: a product recommendation method based on a collaborative filtering model, which specifically includes the following steps:

[0005] Step S1: Collect user historical data;

[0006] Step S2: Preprocess the data;

[0007] Step S3: Collaborative filtering model training;

[0008] Step S4: Perform model validation;

[0009] Step S5: Conduct secondary training to develop recommendations for highly personalized product specifications.

[0010] According to the above technical solution, the method for collecting user historical data in step S1 is as follows: establish a user behavior list, collect the user's historical shopping behavior on the e-commerce platform, including the user's long stay on the product page, clicks, additions to cart, and purchased products; collect the category information, price, and sales trend of the products in the past six months to construct user interest preferences; and collect user basic attributes, product basic attributes, and user scenario information at the same time.

[0011] According to the above technical solution, the method for preprocessing data in step S2 specifically includes: acquiring user historical data, narrowing the data range, deleting outliers, eliminating noise, filling in missing values ​​and deleted outliers, and finally normalizing the data to the scope of digital signal processing; at the same time, further converting the data into matrix form to facilitate model training.

[0012] According to the above technical solution, the method for training the collaborative filtering model in step S3 is as follows:

[0013] Based on specific business scenarios, a collaborative filtering model is used to train the recommendation results. The specific calculation formula in the collaborative filtering model is as follows:

[0014]

[0015] Where i,j represent products, u,v represent users, and W ij This represents the weight factor of the product for i and j; based on the collaborative filtering model calculation formula, the processed data is input into the model for training; after training, the model outputs the similarity between products; in order to reduce the influence of top products, the similarity results between products are weighted in combination with the sales volume of specific products; that is, the similarity between products is obtained based on training based on all user behaviors.

[0016] According to the above technical solution, the value of the weighting factor is determined by the sales volume over the past six months.

[0017] According to the above technical solution, the model verification method is as follows: the model is verified by using the index evaluation method to ensure the rationality and accuracy of the model training results. The index evaluation method mainly uses recall rate and popular product ratio to evaluate the recommendation results.

[0018] According to the above technical solution, step S5 further includes: after outputting the training recommendation product results, obtaining the user behavior list, product category, product basic attributes, and user basic attributes a second time, formulating a product specification training model, and analyzing the specific specifications of the user-recommended products;

[0019] The analysis method is as follows: Using a user behavior list, obtain the products purchased by the user and different specifications of the same product. Establish a comment node for each product, providing basic attributes of different users who have purchased the product and their comments within that node. Record the specifications of the product purchased by different users. Then, establish a chain of related attribute tags centered on this product to obtain the related attribute tag chains for each of the user's purchased product specifications. Repeat the above steps to obtain the related attribute tag chains for all product specifications purchased by the user. Next, based on the current recommended product results, similarly establish related attribute tag chains for different specifications of the recommended products. Finally, match these with the related attribute tag chains for all product specifications purchased by the user, selecting the product specification corresponding to the tag chain with the highest overlap.

[0020] A product recommendation method and system based on a collaborative filtering model, characterized in that: the system includes:

[0021] The data collection module is used to collect user and product data in e-commerce platform scenarios;

[0022] The data preprocessing module is used to preprocess the collected data;

[0023] The collaborative model training module is used to train recommendation results based on the collaborative filtering model of user behavior;

[0024] The model validation module is used to verify whether its recommendation results match user behavior trends;

[0025] The deep customization module is used for secondary training to develop recommendations for highly personalized product specifications.

[0026] Compared with existing technologies, the beneficial effects achieved by this invention are as follows: This invention collects user behavior data, product data, etc., and uses data mining and machine learning techniques to analyze the data, thereby extracting user characteristics, product characteristics, and scenario characteristics, establishing a user-specific interest and preference model, and finally finding products that users are interested in from a massive number of products. At the same time, it formulates recommendations with deeply personalized product specifications, further increasing users' interest in recommended products, meeting users' personalized needs, improving user experience, and helping enterprises to discover long-tail products, achieve balanced product traffic on the platform, and improve the conversion efficiency of existing products. Attached Figure Description

[0027] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0028] Figure 1The flowchart illustrates a product recommendation method based on a collaborative filtering model provided in Embodiment 1 of the present invention.

[0029] Figure 2 This is a schematic diagram of the module composition of a product recommendation system based on a collaborative filtering model, provided in Embodiment 2 of the present invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] Example 1, Figure 1 This is a flowchart of a product recommendation method based on a collaborative filtering model provided in Embodiment 1 of the present invention. This embodiment is applicable to personalized product recommendations during e-commerce shopping. The method can be executed by a product recommendation system based on a collaborative filtering model provided in this embodiment of the present invention. Figure 1 As shown, the method specifically includes the following steps:

[0032] S1. Collect user historical data;

[0033] In this embodiment of the invention, a user behavior list is established to collect users' historical shopping behavior on e-commerce platforms, including prolonged stays on product pages, clicks, adding items to cart, and purchased items. Information such as product category, price, and sales trends over the past six months are also collected to construct user interest preferences. Additionally, basic user attributes, basic product attributes, and user scenario information are collected.

[0034] S2. Preprocess the data;

[0035] In this embodiment of the invention, the collected information is preprocessed. The preprocessing method includes: acquiring historical user data, narrowing the data range, deleting outliers, eliminating noise, and filling in missing and deleted outliers. Finally, the data is normalized to the range of digital signal processing. Simultaneously, the data is further converted into matrix form to facilitate model training. In this embodiment, the data mainly originates from event tracking, which has relatively high noise levels. To best fit user interests and preferences, data preprocessing is necessary before model training. The processing is deeply integrated with specific business scenarios. In the user behavior list, although prolonged stays on product pages, clicks, adding to cart, and purchases are all positive user behaviors, prolonged stays and clicks have a certain degree of randomness. Therefore, different weights are set to differentiate them. When calculating recent sales of products, appropriate time decay is applied to top-selling products to reduce their impact.

[0036] S3, Collaborative Filtering Model Training;

[0037] In this embodiment of the invention, in conjunction with a specific business scenario, a collaborative filtering model is used to train the recommendation results. The specific calculation formula in the collaborative filtering model is as follows:

[0038]

[0039] Where i,j represent products, u,v represent users, and W ij This represents the weight factor of product i, j. Based on the collaborative filtering model's calculation formula, the processed data is input into the model for training. After training, the model outputs the similarity between products. To reduce the influence of top-performing products, the similarity results are weighted by considering the sales volume of specific products. In other words, the similarity between products is obtained based on training across all user behaviors.

[0040] For example, when multiple users are interested in two specific products, the analysis determines that the two products have a certain degree of similarity. The similarity between products is not conveyed through basic product attributes, but rather established through user relationships. Furthermore, in the e-commerce sales field, there are always best-selling and relatively unpopular products. To reduce the influence of top-selling products and give less popular products more exposure, collaborative filtering models de-weight popular product pairs based on their sales volume over the past six months.

[0041] S4. Perform model validation.

[0042] In this embodiment of the invention, after the model training is completed, a metrics evaluation method is used to verify the model, thereby ensuring the rationality and accuracy of the model training results. The metrics evaluation method mainly uses indicators such as recall rate and the proportion of popular products to evaluate the recommendation results, and employs a preview method, i.e., inputting product i to check whether its recommendation list is reasonable and whether it conforms to the behavioral trends of most users. This embodiment of the invention mainly applies two business scenarios: product details and shopping cart. Their relative improvement effect can be verified through the results in Table 1-1. Here, conversion rate refers to the conversion from exposure UV to actual successful payment UV; average order value refers to the ratio of actual revenue to the number of orders; and average revenue per user can measure user value, referring to the average spending power per user among exposure UVs.

[0043] Scene Conversion rate Customer order per capita income Details 19.6％ 1.9％ 22.3％ shopping cart 17.8％ 1.4％ 18.4％

[0044] Table 1-1

[0045] For example, the main metrics include the proportion of popular products in the recommendation list, recall rate, etc. In addition, by combining the user's specific recommendation list, it is also verified whether the recommendation results are consistent with the user's behavioral trends, whether the recommended products are out of stock, or whether the sizes are incomplete, etc.

[0046] S5. Conduct secondary training to develop recommendations for highly personalized product specifications.

[0047] For example, in this embodiment of the invention, after outputting the training recommendation product results, the user behavior list, product category, product basic attributes, and user basic attributes are obtained a second time. A product specification training model is then developed to analyze the specific specifications of the user-recommended products. The analysis method is as follows: Using the user behavior list, the products purchased by the user and different specification options for the same product are obtained. A comment node is established for each single product. Within this node, the basic attributes of different users who have purchased the product and their comments in the comment node are provided. The specifications of the product purchased by different users are recorded. Then, a related attribute tag chain is established around this product to obtain the related attribute tag chain under the user's purchased product specifications. The above analysis steps are then repeated to obtain the related attribute tag chains under all product specifications purchased by the user. Next, based on the current recommended product results, the same process of establishing related attribute tag chains for different specifications of the recommended products is performed. Finally, these are matched with the related attribute tag chains under all product specifications purchased by the user, and the product specification corresponding to the tag chain with the highest overlap is selected.

[0048] For example, after analyzing and recommending a long dress, if the user's historical purchase history reveals tags such as "composed," "reserved," and "business-like," then the long dress is further differentiated by color and size, including highly saturated colors, floral patterns, and black. The black version best matches the "composed," "reserved," and "business-like" tags compared to other colors. After secondary training, personalized product recommendations are further developed. When recommending this dress on the e-commerce platform, the black version is displayed on the product's external page. This increases user interest in the recommended product, improves shopping efficiency, meets personalized needs, and enhances user experience. Simultaneously, it helps businesses discover long-tail products, achieve balanced platform traffic, and improve existing customer conversion rates.

[0049] Example 2: This invention provides a product recommendation system based on a collaborative filtering model. Figure 2 This is a schematic diagram of the module composition of a product recommendation system based on a collaborative filtering model provided in Embodiment 2 of the present invention, as shown below. Figure 2 As shown, the system includes:

[0050] The data collection module is used to collect user and product data in e-commerce platform scenarios;

[0051] The data preprocessing module is used to preprocess the collected data;

[0052] The collaborative model training module is used to train recommendation results based on the collaborative filtering model of user behavior;

[0053] The model validation module is used to verify whether its recommendation results match user behavior trends;

[0054] The deep customization module is used for secondary training to develop recommendations for highly personalized product specifications.

[0055] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0056] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A product recommendation method based on a collaborative filtering model, characterized in that: The method specifically includes the following steps: Step S1: Collect user historical data; Step S2: Preprocess the data; Step S3: Collaborative filtering model training; Step S4: Perform model validation; Step S5: Conduct secondary training to develop recommendations for highly personalized product specifications; Step S5 further includes: after outputting the training recommendation product results, obtaining the user behavior list, product category, product basic attributes, and user basic attributes a second time, formulating a product specification training model, and analyzing the specific specifications of the user-recommended products; The analysis method is as follows: Using a user behavior list, obtain the products purchased by the user and different specifications of the same product. Establish a comment node for each product, providing basic attributes of different users who have purchased the product and their comments within that node. Record the specifications of the product purchased by different users. Then, establish a chain of related attribute tags centered on this product to obtain the related attribute tag chains for each of the user's purchased product specifications. Repeat the above steps to obtain the related attribute tag chains for all product specifications purchased by the user. Next, based on the current recommended product results, similarly establish related attribute tag chains for different specifications of the recommended products. Finally, match these with the related attribute tag chains for all product specifications purchased by the user, selecting the product specification corresponding to the tag chain with the highest overlap.

2. The product recommendation method based on a collaborative filtering model according to claim 1, characterized in that: The method for collecting user historical data in step S1 is as follows: establish a user behavior list and collect the user's historical shopping behavior on the e-commerce platform, including the user's long-term stay on the product page, clicks, additions to cart, and purchased products. Collect information on the product category, price, and sales trends over the past six months to construct user interests and preferences; at the same time, collect basic user attributes, basic product attributes, and user scenario information.

3. The product recommendation method based on a collaborative filtering model according to claim 1, characterized in that: The method for preprocessing data in step S2 specifically includes: acquiring user historical data, narrowing the data range, deleting outliers, eliminating noise, filling in missing values ​​and deleted outliers, and finally normalizing the data to the range of digital signal processing; at the same time, the data is further converted into matrix form to facilitate model training.

4. The product recommendation method based on a collaborative filtering model according to claim 1, characterized in that: In step S3, the method for training the collaborative filtering model is as follows: Based on specific business scenarios, a collaborative filtering model is used to train the recommendation results. The specific calculation formula in the collaborative filtering model is as follows: ； Where i,j represent products, and u,v represent users. This represents the weight factor of the product for i and j; based on the collaborative filtering model calculation formula, the processed data is input into the model for training; after training, the model outputs the similarity between products; in order to reduce the influence of top products, the similarity results between products are weighted in combination with the sales volume of specific products; that is, the similarity between products is obtained based on training based on all user behaviors.

5. The product recommendation method based on a collaborative filtering model according to claim 4, characterized in that: The weighting factor is determined based on sales volume over the past six months.

6. The product recommendation method based on a collaborative filtering model according to claim 1, characterized in that: The specific method for model validation is as follows: model validation is performed using an indicator evaluation method to ensure the rationality and accuracy of the model training results. The indicator evaluation method uses recall rate and the proportion of popular products to evaluate the recommendation results.

7. A system for applying the product recommendation method based on a collaborative filtering model as described in any one of claims 1-6, characterized in that: The system includes: The data collection module is used to collect user and product data in e-commerce platform scenarios; The data preprocessing module is used to preprocess the collected data; The collaborative model training module is used to train recommendation results based on the collaborative filtering model of user behavior; The model validation module is used to verify whether its recommendation results match user behavior trends; The deep customization module is used for secondary training to develop recommendations for highly personalized product specifications.

Images