Method for recommending a product

Abstract

The application provides a commodity recommendation method, and relates to the technical field of electronic commerce. The method comprises the following steps: obtaining user information, wherein the user information comprises historical order information, a behavior log and basic portrait information of a target user; obtaining similar users according to the user information; calculating period information of a commodity to be recommended based on historical order information, a consumption period and purchase information of the similar users, wherein the period information comprises group purchase period information and user purchase period information; determining candidate commodities in the commodity to be recommended according to the period information and the behavior log; and recommending the candidate commodities. The embodiment of the application can combine preferences and the repeat purchase period of commodities to recommend commodities, facilitate accurate grasping of a recommendation opportunity, improve the matching degree of a recommendation time and user demand, improve the shopping experience of users, and thus effectively improve the conversion rate of commodities.

Description

Technical Field

[0001] This application relates to the field of e-commerce technology, and more specifically, to a product recommendation method. Background Technology

[0002] When users purchase goods on e-commerce platforms, the platforms typically push relevant products to users based on their browsing and purchasing history. However, different products have different consumption cycles. For example, for fast-moving consumer goods such as daily necessities and food, users often repurchase similar products when their existing stock is about to run out. Most existing product recommendation methods only match products based on user preferences and cannot accurately grasp the timing of recommendations based on the actual consumption patterns of products. This often results in recommendations being made too early or too late, which not only affects the user's shopping experience but also fails to effectively improve the conversion rate of products. Summary of the Invention

[0003] This application provides a product recommendation method that addresses the problem that existing product recommendation methods rely solely on user preferences, making it difficult to accurately determine the timing of recommendations, thus impacting the user shopping experience and failing to improve product conversion rates. To achieve this objective, this application provides the following solutions.

[0004] According to one aspect of the embodiments of this application, a product recommendation method is provided, including: Obtain user information, including the target user's historical order information, behavior logs, and basic profile information; Based on the user information, similar users are obtained, and the periodic information of the products to be recommended is calculated based on the historical order information and the purchase information of the similar users. The periodic information includes group purchase periodic information and user purchase periodic information. Based on the periodic information and the behavior log, candidate products are determined from the products to be recommended, and the candidate products are recommended.

[0005] In one possible implementation, obtaining similar users based on the user information includes: Based on the user information, the characteristics of the target user are extracted, and the characteristics include at least one of the following: statistical features, spending power features, and behavioral preference features. The similarity between candidate users and target users in the user pool is calculated based on the features, and similar users are selected based on the similarity and the product purchase records of the candidate users and the target users.

[0006] In one possible implementation, the periodicity information of the product to be recommended is calculated based on the historical order information and the purchase information of similar users, including: For each product to be recommended, group purchase cycle information and user purchase cycle information are obtained based on the purchase information of similar users and the historical order information of the target user, respectively. Obtain the number of user purchases corresponding to the product to be recommended, and obtain the period information based on the number of user purchases, the group purchase cycle information, and the user purchase cycle information. The period information includes the dynamic fusion cycle and the cycle confidence score.

[0007] In one possible implementation, the group purchase cycle information includes the group's average repurchase cycle and the group's cycle standard deviation. Obtaining the group purchase cycle information includes: The average repurchase cycle of the group and the standard deviation of the group's cycle are calculated using a first preset calculation formula and the purchase information of the similar users. The first preset calculation formula includes:

[0008]

[0009] ; In the formula, Let n be the average repurchase cycle of the i-th similar user, and n be the number of purchases made by the i-th similar user. Let m be the time of the k-th purchase, and m-1 be the total number of similar users. The average repurchase cycle for the group. The standard deviation of the group period.

[0010] In one possible implementation, the user purchase cycle information includes the individual's average repurchase cycle and the individual's cycle standard deviation. Obtaining the user purchase cycle information includes: The user's purchase cycle information is calculated using a second preset calculation formula and the historical order information. The second preset calculation formula includes: ; ; In the formula, This represents the average repurchase cycle for an individual. For single purchase intervals, , Let p be the individual periodic standard deviation, and p be the number of purchases made by the target user.

[0011] In one possible implementation, obtaining a dynamic fusion period based on the user's purchase frequency, the group purchase cycle information, and the user purchase cycle information includes: Obtain the dynamic weight coefficient corresponding to the number of times the user made a purchase; The dynamic fusion cycle is calculated based on the dynamic weighting coefficient, the average repurchase cycle for individuals, and the average repurchase cycle for groups.

[0012] In one possible implementation, obtaining the periodic confidence score includes: The period confidence score is calculated based on a third preset formula, which is: =Basic periodicity × Stability coefficient × Habit coefficient In the formula, For the period confidence score, the base period score = , The dynamic fusion cycle is represented by w, where w is the range of cycle fluctuations. The stability coefficient is the difference between the current time and the last purchase time. It is determined based on the number of user purchases, the individual periodic standard deviation, and the group periodic standard deviation. The habit coefficient is determined based on the degree of deviation of the user's purchase frequency from the group average level.

[0013] In one possible implementation, determining candidate products among the products to be recommended based on the periodic information and the behavior log includes: Calculate the target user's behavioral intent score towards the recommended products based on the behavioral logs; The recommendation score for the product to be recommended is calculated based on the behavioral intent score and the periodic confidence score. Candidate products are selected from the products to be recommended based on the recommended score.

[0014] In one possible implementation, the recommendation score of the product to be recommended is calculated based on the behavioral intent score and the periodic confidence score, including: The restocking period information of the product to be recommended is obtained based on the dynamic fusion cycle and the time of the last purchase of the product to be recommended. The recommendation score is calculated based on the replenishment period information, the behavioral intent score, and the periodic confidence score.

[0015] In one possible implementation, the recommendation score is calculated as follows:

[0016] In the formula, For recommended score, The periodic confidence score. The weights for the periodic confidence scores. Score the behavioral intent. Weights are assigned to the behavioral intent.

[0017] The beneficial effects of the technical solutions provided in this application are: The product recommendation method provided in this application includes: obtaining user information, including the target user's historical order information, behavior logs, and basic profile information; obtaining similar users based on the user information; calculating the periodic information of the products to be recommended based on historical order information, consumption cycles, and purchase information of similar users; the periodic information includes group purchase cycle information and user purchase cycle information; determining candidate products among the products to be recommended based on the periodic information and behavior logs; and recommending candidate products. This application's embodiments can combine preferences and product repurchase cycles for product recommendations, facilitating accurate timing of recommendations, improving the matching degree between recommendation time and user needs, enhancing the user's shopping experience, and thus effectively increasing product conversion rates. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below.

[0019] Figure 1 A flowchart of the product recommendation method provided in the embodiments of this application; Figure 2 This is a flowchart illustrating the execution of a product delivery method provided in an embodiment of this application. Detailed Implementation

[0020] The embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the embodiments described below with reference to the accompanying drawings are exemplary descriptions for explaining the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions of the embodiments of this application.

[0021] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the terms “comprising” and “including” as used in embodiments of this application mean that the corresponding feature can be implemented as the presented feature, information, data, step, operation, element, and / or component, but do not exclude implementation as other features, information, data, step, operation, element, component, and / or combinations thereof supported by the art. It should be understood that when we say that an element is “connected” or “coupled” to another element, the one element can be directly connected or coupled to the other element, or it can mean that the one element and the other element establish a connection relationship through an intermediate element. Furthermore, “connected” or “coupled” as used herein can include wireless connection or wireless coupling. The term “and / or” as used herein indicates at least one of the items defined by the term; for example, “A and / or B” indicates implementation as “A,” or implementation as “A,” or implementation as “A and B.”

[0022] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0023] The technical solutions of this application and their effects are described below through several exemplary embodiments. It should be noted that the following embodiments can be referenced, borrowed from, or combined with each other. Identical terms, similar features, and similar implementation steps in different embodiments will not be repeated.

[0024] It should be noted that the collection, storage, use, processing, transmission, provision, and disclosure of data or user information involved in the technical solution of this application all comply with relevant laws and regulations and do not violate public order and good morals. The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, and displayed data) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use, and processing of related data must comply with relevant laws, regulations, and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse. In addition, certain software, components, models, and other existing industry solutions may be mentioned in the embodiments of this application. These should be considered exemplary, and their purpose is merely to illustrate the feasibility of implementing the technical solution of this application, but does not mean that the applicant has already used or necessarily used such solutions.

[0025] The product recommendation method provided in this application aims to solve at least one technical problem existing in the prior art.

[0026] Optionally, the product recommendation method of this application can be used on mobile phones, computers, servers, cloud platforms, and other terminals capable of recommending products when users use the device.

[0027] Optionally, the terminal may include a data acquisition module, a similar user clustering engine, a periodic calculation engine, a behavior analysis engine, and a recommendation generation module. The data acquisition module extracts users' historical order information, behavior logs, and user profiles (basic profile information). The similar user clustering engine calculates similarity based on user profiles and behavior logs to retrieve similar users. The periodic calculation engine includes individual periodicity submodules, group periodicity submodules, and dynamic fusion submodules, which obtain periodicity information. The behavior analysis engine calculates behavioral intent scores. The recommendation generation module integrates behavioral intent scores and periodicity information to generate a ranked list of candidate products.

[0028] Options, such as Figure 1 , Figure 2 As shown, the product recommendation method of this application includes: S101: Obtain user information.

[0029] Optionally, user information can be obtained from the e-commerce platforms used by the target users. This user information may include the target users' historical order information, behavior logs, and basic profile information. The basic profile information may include the target users' age, region, and consumption level, while the behavior logs may include product-related behaviors such as searching for products, adding items to cart, viewing details, and browsing.

[0030] Optionally, when recommending products, the consumption cycle of the products to be recommended can be obtained first. The acquisition of the consumption cycle includes: determining the type of the product to be recommended based on the product attributes, including product category and specifications, and the type including fast-moving consumer goods, slow-moving consumer goods, and durable goods; and obtaining the consumption cycle of the product to be recommended based on the type.

[0031] Optionally, products to be recommended can be obtained based on the recommendation requirements of the e-commerce platform, or products to be recommended can be determined based on received recommendation instructions. Products to be recommended can also be determined based on information such as the target user's login time, login location, and login device.

[0032] Optionally, different consumption cycles can be set for different types of goods. The classification information of different goods can be pre-stored. After obtaining the information of the goods to be recommended, the classification of the goods to be recommended can be determined based on the classification information.

[0033] Optionally, the consumption cycle of fast-moving consumer goods (FMCG) is shorter than that of slow-moving consumer goods (SMO), and the consumption cycle of SMO is shorter than that of durable goods.

[0034] In one embodiment, the consumption cycle of fast-moving consumer goods (FMCG) is ≤30 days (e.g., food, paper towels), and the consumption cycle of slow-moving consumer goods (SMO) is ≤180 days (e.g., household appliance supplies, skincare products). The consumption cycle of durable goods is >180 days (e.g., appliances, furniture). The consumption cycles of recommended products of the same type can be the same or different, and the value of the consumption cycle can be determined based on actual needs or the purchase frequency and usage habits of the recommended products (e.g., large-scale use on specific dates or in specific regions).

[0035] Optionally, user information can be collected each time a user uses a terminal or an e-commerce platform or website, and user information can be extracted based on this information. After obtaining user information, it can be stored in a database, where historical order information, behavior logs, basic profile information, and other information can be stored in the database as fields.

[0036] S102: Obtain similar users based on user information, and calculate the cycle information of the products to be recommended based on historical order information and the purchase information of similar users.

[0037] Optionally, obtaining similar users based on user information includes: extracting features of the target user based on user information, including at least one of manual statistical features, spending power features, and behavioral preference features; calculating the similarity between candidate users in the user pool and the target user based on the features; and filtering similar users based on the similarity and the purchase records of the candidate users and the target user.

[0038] Optionally, demographic features can be directly read from basic profile information, consumption capacity features can be statistically calculated based on historical order information, and behavioral preference features can be obtained from behavior logs and historical order information.

[0039] In one embodiment, demographic characteristics may include age, gender, and region; spending power characteristics may include average order value and total spending; and behavioral preference characteristics may include preferred product category distribution and active time periods. After extracting the characteristics of the target user, these characteristics can be transformed into feature vectors to facilitate the acquisition of similar users.

[0040] Optionally, each user's information can be placed into a user pool, and similar users of the target user can be found based on the user pool.

[0041] Alternatively, the similarity can be calculated using the following formula: In the formula, This represents the similarity between user u and user i. Let i be the feature vector of user i. Let be the feature vector of user u. After obtaining the similarity between the target user and other users (candidate users) in the user pool, a predetermined number of candidate users with the highest similarity (e.g., 50, or other numbers, the specific number can be determined according to actual needs) can be selected as similar users. Specifically, the K users with the highest similarity (e.g., K=50) can be selected to form a group set G of similar users.

[0042] Optionally, to improve the similarity between similar users and the target user, a similarity coefficient can be calculated between candidate users in the user pool and the target user. Based on this similarity coefficient, the similarity is weighted (e.g., by multiplying the similarity by the similarity coefficient, or by adding the number corresponding to the similarity coefficient to the similarity). The candidate users with the highest weighted values ​​are then selected as similar users. The similarity coefficient can be the Jaccard similarity coefficient, and its calculation formula is as follows: A represents the set of purchased items by the target user, and B represents the set of purchased items by the candidate user. It refers to the quantity of goods purchased jointly. It represents the total number of duplicate items purchased by the two individuals.

[0043] Optionally, a first candidate user with a similarity greater than a predetermined threshold (e.g., 70%) can be selected from the user pool based on the characteristics of the target user. A second candidate user with an overlap greater than a predetermined value can be selected based on the overlap of the products purchased by the candidate user and the target user (which can be represented by the Jaccard similarity coefficient or other similarity coefficients). The object that belongs to both the first candidate user and the second candidate user is identified as a similar user and placed into the group set G.

[0044] Alternatively, only similarity statistics can be performed, and all objects in the user pool whose similarity to the target user is greater than a predetermined similarity threshold can be considered as similar users.

[0045] Optionally, in scenarios where user information indicates that a user rarely purchases goods (e.g., purchases less than 3 times), has no purchase information, or has no e-commerce platform usage information, it is not necessary to search for similar users. Instead, the average repurchase cycle of the entire platform for the category of the product to be recommended can be used as the average repurchase cycle of similar users for subsequent calculations.

[0046] Optionally, the periodicity information of the products to be recommended is calculated based on historical order information and purchase information of similar users. This includes: for each product to be recommended, obtaining group purchase periodicity information and user purchase periodicity information based on the purchase information of similar users and the historical order information of the target users; obtaining the number of times users purchase the product to be recommended (e.g., when the product to be recommended is garbage bags, obtaining the number of times the target user has purchased various brands of garbage bags based on historical order information); and obtaining periodicity information based on the number of user purchases, group purchase periodicity information, and user purchase periodicity information. The periodicity information includes a dynamic fusion periodicity and a periodicity confidence score. The purchase information of similar users may include their historical product purchase information.

[0047] Optionally, the group purchase cycle information includes the group's average repurchase cycle and the group's cycle standard deviation. Obtaining the group purchase cycle information includes: calculating the group's average repurchase cycle and the group's cycle standard deviation using a first preset calculation formula and purchase information of similar users. The first preset calculation formula includes: ; ; ; In the formula, Let n be the average repurchase cycle of the i-th similar user, and n be the number of purchases made by the i-th similar user. For the time of the kth purchase, The average repurchase cycle for the group represents the typical repurchase frequency of a group with characteristics highly similar to the target users for this product, where m-1 is the total number of similar users. The standard deviation of the group period is used to calculate the group purchase period information for each product to be recommended.

[0048] Optionally, the user purchase cycle information includes the individual's average repurchase cycle and the individual's cycle standard deviation. Obtaining the user purchase cycle information includes: calculating the user purchase cycle information using a second preset calculation formula and historical order information. The second preset calculation formula includes: ; ; In the formula, This represents the average repurchase cycle for an individual. For single purchase intervals, , Let p be the individual periodic standard deviation, and p be the number of purchases by the target user. This formula can be used to obtain the user's purchase period information for each product to be recommended. When the product to be recommended is a high-frequency purchase item (e.g., when the average number of purchases by all users is greater than 2), p can also be compared with the average number of purchases. If p is equal to or less than the average number of purchases, then... If p is 0, then p = 0. .

[0049] Optionally, for a product to be recommended, if the target user has made fewer than 2 purchases, the average repurchase cycle for the individual is deemed unavailable, and the calculation of the user's purchase cycle information is stopped.

[0050] Optionally, a dynamic fusion period is obtained based on user purchase frequency, group purchase cycle information, and user purchase cycle information, including: obtaining the dynamic weight coefficient corresponding to the user purchase frequency; and calculating the dynamic fusion period based on the dynamic weight coefficient, the individual average repurchase cycle, and the group average repurchase cycle.

[0051] Optionally, the scenario corresponding to the number of user purchases can be obtained, and a dynamic weighting coefficient can be determined based on the scenario. Furthermore, different dynamic fusion cycle calculation formulas can be set for different scenarios, and the dynamic fusion cycle can be calculated according to these formulas.

[0052] Optionally, the dynamic weighting coefficient can be represented as λ, 0≤λ≤1, and the magnitude of this coefficient is positively correlated with the number of times a user makes a purchase.

[0053] In one embodiment, the scenario may include a cold start / sparse scenario, a growth phase scenario, and a maturity phase scenario. Specifically, the cold start / sparse scenario corresponds to the case where the number of user purchases p < 2, in which case λ = 0. In this case, the group data of similar users is fully trusted, and the dynamic fusion cycle is calculated using the following formula: = , This is a dynamic fusion cycle.

[0054] When 2 ≤ number of user purchases p ≤ 5, corresponding to the growth stage scenario, λ increases linearly with p (e.g., λ = 0.2 × (p)). 1) The formula for calculating the dynamic fusion cycle is: .

[0055] When the number of user purchases p > 5, corresponding to the mature stage, λ ≈ 0.9. At this point, individual data is primarily trusted, but 10% of group data is retained for smoothing anomalies. Therefore, the formula for calculating the dynamic fusion cycle can be: As personal data accumulates, the system gradually transitions from "collective intelligence" to "personal habits," which not only solves the cold start problem but also allows the use of collective data to correct occasional individual anomalies, thereby improving data accuracy.

[0056] Optionally, when calculating the dynamic fusion period, the dynamic fusion period can be set to the average repurchase period of the group when the user has made 1 purchase. When the user has made 2-4 purchases, λ can be set to 0.5. When a user makes 5 or more purchases, set λ=0.9. .

[0057] Alternatively, when calculating the dynamic fusion cycle, the average repurchase cycle of the group can be used as the prior probability distribution, and the average repurchase cycle of the individual can be used as the likelihood evidence. The posterior probability distribution can be calculated using Bayes' formula to determine the final dynamic fusion cycle, rather than a simple linear weighting.

[0058] Optionally, obtaining the periodic confidence score includes: calculating the periodic confidence score based on a third preset formula, whereby the third preset formula is: =Basic periodicity × Stability coefficient × Habit coefficient In the formula, For the period confidence score, the base period score = , The dynamic fusion cycle is represented by w, where w is the range of cycle fluctuations. The stability coefficient is the difference between the current time and the last purchase time (the time when the recommended product was last purchased). The stability coefficient is determined based on the number of user purchases, the individual periodic standard deviation, and the group periodic standard deviation. The habit coefficient is determined based on the degree of deviation of the user's purchase frequency from the group average level.

[0059] Optionally, the periodic fluctuation range can be calculated based on the dynamic fusion cycle. Specifically, the formula for calculating the periodic fluctuation range can be: This also allows you to obtain the consumption cycle of the product to be recommended, and multiply that consumption cycle by a predetermined value (such as 20% or other values) to obtain the value of the cycle fluctuation range.

[0060] Optionally, when the number of purchases by a user is no greater than a preset value (e.g., 2), the stability coefficient can be calculated using the group periodic standard deviation (e.g., the stability coefficient equals the group periodic standard deviation or the group periodic standard deviation is normalized to [0,1], and the normalized result is determined as the stability coefficient). When the number of purchases by a user is greater than the preset value, the stability coefficient can be calculated using the individual periodic standard deviation (the calculation method is the same as that using the group periodic standard deviation).

[0061] Optionally, the value of the habituality coefficient can be limited to a preset range (e.g., 0.8-1.2), and the formula for calculating the habituality coefficient can be:

[0062] In the formula, This is the habitual coefficient. Alternatively, other formulas that can represent the degree of deviation can also be used to calculate the habitual coefficient.

[0063] Alternatively, a Gaussian distribution probability can be calculated using a dynamic fusion period, and this probability can be used as the period confidence score. Another approach is to use the individual average repurchase period, the group average repurchase period, the standard deviation (individual period standard deviation, group period standard deviation), the period confidence score, and historical information about the product to be recommended. Based on this historical information, a machine learning model (such as GBDT) can be trained. After obtaining group purchase period information and user purchase period information, this information, along with the name of the product to be recommended, is input into the machine learning model to obtain the period confidence score output by the model. This score is then used as the period confidence score for the product to be recommended.

[0064] S103: Based on periodic information and behavior logs, determine the candidate products among the products to be recommended, and recommend the candidate products.

[0065] Optionally, candidate products are determined from the products to be recommended based on periodic information and behavior logs, including: calculating the target user's behavioral intent score for the products to be recommended based on the behavior logs; calculating the recommendation score for the products to be recommended based on the behavioral intent score and the periodic confidence score; and filtering candidate products from the products to be recommended based on the recommendation score.

[0066] Optionally, a user's recent (e.g., one-month) behavioral sequence can be obtained from behavioral logs, and the user's purchase intent strength (i.e., behavioral intent score) for each recommended product can be calculated based on this behavioral sequence. Specifically, the formula for calculating this behavioral intent score can be:

[0067] This allows users to preset the weight of a specific behavior within a sequence of actions. Specifically, the weight for searching can be 1.0, adding to cart 0.8, viewing details 0.7, and browsing 0.3. The value is the number of days since the i-th action in the behavior sequence (or the number of days since the last action). The count can be the number of times a certain action in the behavior sequence has been executed. The scores of each action corresponding to the product to be recommended are summed to obtain the behavior intent score.

[0068] Optionally, when calculating the behavioral intent score, negative behavior deductions (such as canceling an order or removing it from the shopping cart) can be introduced. The calculated behavioral intent score is subtracted based on these negative behaviors, or negative weights are assigned to the negative behaviors. In particular, if there are strong negative behaviors (such as putting the item corresponding to the product to be recommended on a second-hand platform), the intent score is directly set to zero.

[0069] Alternatively, a time-based scoring method can be used to obtain behavioral intent scores (e.g., 100 points for viewing the product within 1 hour, 60 points for viewing it within 1 day, etc.).

[0070] Optionally, the recommendation score of the product to be recommended is calculated based on the behavioral intent score and the periodic confidence score, including: obtaining the restocking period information of the product to be recommended based on the dynamic fusion period and the time of the last purchase of the product to be recommended; and calculating the recommendation score based on the restocking period information, the behavioral intent score, and the periodic confidence score.

[0071] Optionally, replenishment period information may include whether the product is close to its replenishment period. The weights of the behavioral intent score and the periodic confidence score can be adjusted based on the replenishment period information, and the recommendation score can be calculated based on the adjusted weights.

[0072] Alternatively, the recommendation score can be calculated as follows:

[0073] In the formula, For recommended score, The weights for the periodic confidence scores. Weights are assigned to the behavioral intent.

[0074] Optionally, different replenishment periods can be set according to the type of product to be recommended. For example, the replenishment period for fast-moving consumer goods (FMCG) can be shorter than that for slow-moving consumer goods (SMO), and the replenishment period for SMO can be shorter than that for durable goods. Alternatively, the replenishment period can be set to a fixed value.

[0075] In one embodiment, the weight adjustment rule can be: If it is close to the replenishment period (i.e.) If the period is ≤5 days, then α=0.7, β=0.3 (period confidence score dominates).

[0076] If it is far from the replenishment period (i.e.) If the number of days is greater than 15 days, then α = 0.3, β = 0.7 (behavioral intention is dominant, capturing sudden needs).

[0077] If in the intermediate state (5 days < If the number of days is ≤15 days, then α=0.5, β=0.5.

[0078] because The correction cycle has been integrated with similar user data, so even during a cold start, it can accurately determine whether it is "close to the replenishment period".

[0079] Optionally, different judgment periods can be set for fast-moving consumer goods (FMCG), slow-moving consumer goods (SMO), and durable goods (e.g., FMCG can use 5 days and 15 days to judge whether it is close to the replenishment period, in the middle stage, or far from the replenishment period; SMO can use 30 days and 90 days to judge; and durable goods can use 90 days and 180 days to judge), so as to set different recommendation sensitivity thresholds according to the type of goods.

[0080] Optionally, after obtaining the recommendation score for each product to be recommended, the products can be sorted according to the score. Specifically, the products can be sorted in descending order of their scores, and candidate products can be selected based on the sorting results and the product category.

[0081] In one embodiment, the recommended products can be sorted in descending order according to their recommendation scores, and the top N (e.g., 20) can be selected as candidate products. Furthermore, when filtering candidate products, a diversity rule can be applied (that is, no more than M candidate products of the same category).

[0082] Optionally, when implementing the method of this application, the purchasing behavior of the target user can also be detected, and the user information and user purchase cycle information of the target user can be updated based on the purchasing behavior. Furthermore, the updated information can be fed back into the user pool, and used to participate in product recommendations for other users and determine candidate products for the next recommendation, thereby achieving self-learning evolution.

[0083] Compared with existing technologies, this application has the following significant advantages: Completely solves the cold start problem: Even for new users who have only made a purchase once, it can provide accurate predictions of repurchase time based on the patterns of similar user groups, significantly improving the referral conversion rate for new users.

[0084] Strong anti-interference capability: It utilizes group data to smooth out the cyclical prediction bias caused by individual abnormal behavior (such as hoarding), making the recommendation timing more robust.

[0085] Recommendation accuracy has been significantly improved: the dual-dimensional integration captures both "regular needs" (cycles) and "sudden needs" (behaviors), achieving full-scenario coverage.

[0086] Reduce user annoyance: Accurate periodic predictions avoid ineffective push notifications immediately after a user's purchase or during periods of non-demand. Combined with frequency control mechanisms, this significantly improves the user experience.

[0087] Maximizing data utilization: The platform has fully explored the value of its accumulated group behavior data, expanding it from simple "item recommendation" to the field of "time prediction".

[0088] The terms "first," "second," "third," "fourth," "1," "2," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in a sequence other than that shown in the illustrations or text descriptions.

[0089] It should be understood that although arrows indicate various operation steps in the flowcharts of this application's embodiments, the order in which these steps are implemented is not limited to the order indicated by the arrows. Unless explicitly stated herein, in some implementation scenarios of this application's embodiments, the implementation steps in each flowchart can be executed in other orders as required. Furthermore, some or all steps in each flowchart, based on the actual implementation scenario, may include multiple sub-steps or multiple stages. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage can also be executed at different times. In scenarios where execution times differ, the execution order of these sub-steps or stages can be flexibly configured according to requirements, and this application's embodiments do not limit this.

[0090] The above description is only an optional implementation method for some implementation scenarios of this application. It should be noted that for those skilled in the art, other similar implementation methods based on the technical concept of this application without departing from the technical concept of this application also fall within the protection scope of the embodiments of this application.

Claims

1. A product recommendation method characterized by comprising: include: Obtain user information, including the target user's historical order information, behavior logs, and basic profile information; Based on the user information, similar users are obtained, and the periodic information of the products to be recommended is calculated based on the historical order information and the purchase information of the similar users. The periodic information includes group purchase periodic information and user purchase periodic information. Based on the periodic information and the behavior log, candidate products are determined from the products to be recommended, and the candidate products are recommended.

2. The product recommendation method according to claim 1, characterized by, Based on the user information, similar users are obtained, including: Based on the user information, the characteristics of the target user are extracted, and the characteristics include at least one of the following: statistical features, spending power features, and behavioral preference features. The similarity between candidate users and target users in the user pool is calculated based on the features, and similar users are selected based on the similarity and the product purchase records of the candidate users and the target users.

3. The product recommendation method according to claim 1, characterized by, The periodicity information of the products to be recommended is calculated based on the historical order information and the purchase information of similar users, including: For each product to be recommended, group purchase cycle information and user purchase cycle information are obtained based on the purchase information of similar users and the historical order information of the target user, respectively. Obtain the number of user purchases corresponding to the product to be recommended, and obtain the period information based on the number of user purchases, the group purchase cycle information, and the user purchase cycle information. The period information includes the dynamic fusion cycle and the cycle confidence score.

4. The product recommendation method according to claim 3, characterized in that, The group purchase cycle information includes the group's average repurchase cycle and the group's cycle standard deviation. Obtaining the group purchase cycle information includes: The average repurchase cycle of the group and the standard deviation of the group's cycle are calculated using a first preset calculation formula and the purchase information of the similar users. The first preset calculation formula includes: ； In the formula, Let n be the average repurchase cycle of the i-th similar user, and n be the number of purchases made by the i-th similar user. Let m be the time of the k-th purchase, and m-1 be the total number of similar users. The average repurchase cycle for the group. The standard deviation of the group period.

5. The product recommendation method according to claim 4, characterized in that, The user purchase cycle information includes the individual's average repurchase cycle and the individual's cycle standard deviation. The acquisition of the user purchase cycle information includes: The user's purchase cycle information is calculated using a second preset calculation formula and the historical order information. The second preset calculation formula includes: ； ； In the formula, This represents the average repurchase cycle for an individual. For each purchase interval, , Let p be the individual periodic standard deviation, and p be the number of purchases made by the target user.

6. The product recommendation method according to claim 5, characterized in that, Based on the user's purchase frequency, the group purchase cycle information, and the user purchase cycle information, a dynamic fusion cycle is obtained, including: Obtain the dynamic weight coefficient corresponding to the number of times the user made a purchase; The dynamic fusion cycle is calculated based on the dynamic weighting coefficient, the average repurchase cycle for individuals, and the average repurchase cycle for groups.

7. The product recommendation method according to claim 5, characterized in that, The acquisition of the periodic confidence score includes: The period confidence score is calculated based on a third preset formula, which is: =Basic periodicity × Stability coefficient × Habit coefficient In the formula, For the period confidence score, the base period score = , The dynamic fusion cycle is represented by w, where w is the range of cycle fluctuations. The stability coefficient is the difference between the current time and the last purchase time. It is determined based on the number of user purchases, the individual periodic standard deviation, and the group periodic standard deviation. The habit coefficient is determined based on the degree of deviation of the user's purchase frequency from the group average level.

8. The product recommendation method according to claim 3, characterized in that, Based on the periodic information and the behavior log, candidate products are determined from the products to be recommended, including: Calculate the target user's behavioral intent score towards the recommended products based on the behavioral logs; The recommendation score for the product to be recommended is calculated based on the behavioral intent score and the periodic confidence score. Candidate products are selected from the products to be recommended based on the recommended score.

9. The product recommendation method according to claim 8, characterized in that, The recommendation score for the product to be recommended is calculated based on the behavioral intent score and the periodic confidence score, including: The restocking period information of the product to be recommended is obtained based on the dynamic fusion cycle and the time of the last purchase of the product to be recommended. The recommendation score is calculated based on the replenishment period information, the behavioral intent score, and the periodic confidence score.

10. The product recommendation method according to claim 9, characterized in that, The formula for calculating the recommendation score is: In the formula, For recommended score, The periodic confidence score. The weights for the periodic confidence scores. Score the behavioral intent. Weights are assigned to the behavioral intent.

Images