A method and system for recalling recommended goods in a second-hand e-commerce scenario

Abstract

This invention relates to a method and system for recalling recommended products in a second-hand e-commerce scenario. The method includes: acquiring real-time user behavior data; extracting the user's identity identifier and the identity identifier of the object product in the user's current behavior based on the real-time user behavior data; acquiring co-occurring products that form a co-occurring product pair with the object product from a data source; calculating the similarity between the object product and each co-occurring product; and sorting the multiple co-occurring products in descending order of similarity, selecting a predetermined number of co-occurring products at the top of the sorting as the recall products. This invention achieves real-time updates of the similarity of each product by storing the user's real-time behavior and promptly deleting expired user behavior, ensuring that the recommendation results are more relevant to the user's current or recent interests.

Description

Technical Field

[0001] This invention relates to the field of Internet applications, and in particular to a method and system for recalling recommended products in a second-hand e-commerce scenario. Background Technology

[0002] With the development of internet applications and the logistics industry, more and more merchants and consumers tend to complete transactions on e-commerce platforms. Currently, many e-commerce platforms offer product and advertising recommendation services to increase product exposure and help users quickly find their desired items. For example, when a user logs in, the platform retrieves the products the user has historically purchased, browsed, and favorited, calculates the products with the highest similarity to these, and recommends these products to the user when they meet the recommendation criteria. Another example is recommending the most popular products based on the product categories the user is currently browsing. Yet another example is identifying user groups similar to the current target user and recommending products purchased by similar users to the current target user. These recommendation methods generally have good results for first-hand e-commerce platform users, or each has its advantages. However, for second-hand e-commerce platforms, many recommendation methods suitable for first-hand e-commerce platforms do not provide good results. This is because the products sold on second-hand e-commerce platforms are second-hand, with significant differences and non-standard characteristics. Even for the same type of product, the actual condition can vary greatly due to different usage conditions. Therefore, there is an urgent need to provide users of second-hand e-commerce platforms with methods for recommending products. Summary of the Invention

[0003] To address the technical problems existing in the prior art, this invention proposes a method and system for recalling recommended products in the context of second-hand e-commerce, which is used to provide effective recommended products for second-hand e-commerce users.

[0004] To address the aforementioned technical problems, according to one aspect of the present invention, a method for recalling recommended products in a second-hand e-commerce scenario is provided, comprising the following steps: acquiring real-time behavior data of a target user; extracting the user identity identifier of the target user and the object product identity identifier of the target user's current behavior from the real-time behavior data; acquiring co-occurring products that form a co-occurring product pair with the object product from a parameter database based on the target user identity identifier; calculating the similarity between the object product and each co-occurring product respectively; and sorting the multiple co-occurring products in descending order of similarity, and selecting a predetermined number of co-occurring products at the top of the sort as the recalled products.

[0005] According to another aspect of the present invention, the present invention also provides a recommended product recall system in a second-hand e-commerce scenario, including a data acquisition module, a data parsing module, a co-occurring product acquisition module, a similarity calculation module, and a recall module. The data acquisition module is configured to acquire real-time behavioral data of a target user; the data parsing module is connected to the data acquisition module and configured to extract the user's user identity identifier and the object product identity identifier of the target user's current behavior from the real-time behavioral data of the target user; the co-occurring product acquisition module is connected to the data parsing module and configured to acquire co-occurring products that form co-occurring product pairs with the object product from a data source; the similarity calculation module is connected to the co-occurring product acquisition module and configured to calculate the similarity between the object product and each co-occurring product; the recall module is connected to the similarity calculation module and configured to sort multiple co-occurring products in descending order of similarity, and select a preset number of co-occurring products at the top of the sorted list as recall products.

[0006] This invention offers enhanced timeliness compared to traditional Item-CF (item-based collaborative filtering) methods. By storing users' real-time behavior, it enables real-time updates of the similarity of each item and promptly removes expired user behavior, ensuring that the recommendation results are more relevant to the user's current or recent interests. Attached Figure Description

[0007] The preferred embodiments of the present invention will now be described in further detail with reference to the accompanying drawings, wherein:

[0008] Figure 1 This is a partial structural principle diagram of a second-hand e-commerce system provided according to an embodiment of the present invention;

[0009] Figure 2 This is a flowchart of a product recommendation method in a second-hand e-commerce scenario provided by an embodiment of the present invention;

[0010] Figure 3 This is a flowchart of a method for updating various parameter data in a Redis database 400 according to an embodiment of the present invention;

[0011] Figure 4 This is a block diagram illustrating the principle of a recommendation system in a second-hand e-commerce scenario according to an embodiment of the present invention;

[0012] Figure 5 This is a block diagram illustrating the principle of a product recall system in a second-hand e-commerce scenario according to an embodiment of the present invention.

[0013] Figure 6 This is a partial principle block diagram of a product recall system in a second-hand e-commerce scenario provided by an embodiment of the present invention;

[0014] Figure 7 This is a block diagram illustrating the principle of a similarity calculation module according to an embodiment of the present invention; and

[0015] Figure 8 This is a partial principle block diagram of a product recall system in a second-hand e-commerce scenario provided by an embodiment of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0017] In the following detailed description, reference can be made to the accompanying drawings, which form part of this application and illustrate specific embodiments of the present application. In the drawings, similar reference numerals describe substantially similar components in different figures. Specific embodiments of the present application are described in sufficient detail below to enable those skilled in the art to implement the technical solutions of the present application. It should be understood that other embodiments may also be utilized, or structural, logical, or electrical changes may be made to the embodiments of the present application.

[0018] This invention provides a method and system for recalling recommended products in a second-hand e-commerce scenario, used to recall qualified products when recommending products to users on a second-hand e-commerce platform. Due to the highly non-standard nature of products in the second-hand e-commerce scenario and the significant differences between individuals, experience has shown that different products clicked by a user within a suitable time period exhibit a certain degree of similarity. For example, when a user browses and views products on a second-hand platform due to a purchase intention, the products clicked within a certain time period generally satisfy the same purchase intention; these products clicked by the same user represent the user's interests. Since click behavior can express user interests, if two products are clicked by many users, it indicates that many users are simultaneously interested in these two products, suggesting that these two products can similarly satisfy the interests of many users, thus indicating their similarity. This invention utilizes this principle, calculating the similarity between two products by statistically analyzing the number of times each product is clicked by the same user and the number of times each product is clicked individually, using this as a product recall strategy. In the following description, multiple products clicked by the same user are paired to form co-occurring product pairs (product 1, product 2), where product 1 and product 2 are co-occurring products.

[0019] Figure 1 This is a partial structural principle diagram of a second-hand e-commerce system according to an embodiment of the present invention. In this embodiment, the second-hand e-commerce system includes a second-hand e-commerce platform 100, where sellers can publish products, and buyers can browse products, place orders, make payments, and post comments. The second-hand e-commerce platform 100 can be implemented as an online website platform and / or an SC system consisting of a server and a client APP. The second-hand e-commerce platform 100 is connected to a Kafka real-time data source 300 and a Redis database 400. Various processes within the second-hand e-commerce platform 100 connect to the Kafka real-time data source 300 through a producer API, storing real-time messages of the processes in the corresponding partitions under the corresponding topics in the Kafka real-time data source 300. Any message published to a partition is appended to the end of the log file. The message format stored in Kafka varies slightly depending on the Kafka version, but generally, messages stored in Kafka include a header and a message body, where the header includes an offset and a message length. The offset is used to uniquely identify a message, indicating its position within the file. The message length records the size of the message, expressed in bytes. The message body (Record) includes multiple fields, such as a CRC32 checksum field, a message format version number field, a message attribute field, and a message body (Value) field. In this embodiment, the message body field may also include multiple fields for storing user token (user identification), product info_id (product identification), product secondary category, user behavior timestamp, user behavior fields, etc. The user behavior fields include various user actions on the platform, such as clicking on products, placing orders, adding to favorites, making payments, and leaving comments.

[0020] Redis Database 400 is a high-performance key-value database. In this embodiment, it stores and maintains various parameter data required by the recommendation system 200 for calculations when recalling products. In one embodiment, Redis Database 400 stores the following information: historical click behavior (such as browsing and favorites) of all platform users within 3 days and their corresponding products, the number of times products were clicked within 3 days, the list of co-occurring products within 3 days, and the number of times each co-occurring product pair was co-occurred. The 3-day period is one embodiment; it can also be 2 days, 1 day, 12 hours, etc. Since user behavior and interests are negatively correlated over time—that is, behavior closer to real-time is more representative of user interests, and behavior from earlier times may no longer represent user interests—a more reliable time window is needed to make the recalled recommended products more reliable. According to experimental results, a time window length of 1-3 days can achieve better recommendation results. In addition, the recommendation system 200 temporarily stores the calculated set of similar products in the Redis database 400 when no products currently meet the recommendation criteria. When products meet the criteria, the recommendation system 200 retrieves them from the Redis database and pushes them to the second-hand e-commerce platform 100 for display to the target user. The recommendation system 200 is connected to the second-hand e-commerce platform 100, the Kafka real-time data source 300, and the Redis database 400, respectively, to monitor user behavior on the second-hand e-commerce platform 100, calculate similar products for the user, and push them to the user when necessary.

[0021] Figure 2 This is a flowchart of a product recommendation method in a second-hand e-commerce scenario according to an embodiment of the present invention. Figure 2 and Figure 1 The explanation is as follows:

[0022] Step S1a involves monitoring user activity on the platform. This includes monitoring actions such as logging into the platform, entering the main page, and browsing products in the product list. Typically, user activity on the platform triggers the product recommendation process.

[0023] Step S2a: Obtain real-time behavioral data of the target user. For example, see... Figure 1 As shown, when the recommendation system 200 connects to the Kafka data source 300, it extracts real-time behavioral messages for the target user from the Kafka data source through the consumer API via the Spark Streaming real-time task deployed to it.

[0024] Step S3a: Parse the message body to obtain the user's identity identifier and the identity identifier of the product involved in the user's behavior. Specifically, the user's behavior type can be obtained from the user behavior field in the message body. When the behavior type is a click, the user token and product info_id can be parsed from the message body, resulting in a data entry containing (user, product). For simplicity, "user" represents the identity identifier of the current target user, i.e., the aforementioned user token, and "product" represents the identity identifier of the clicked product in the user's behavior, such as the aforementioned product info_id.

[0025] Step S4a: Obtain the co-occurring items based on the obtained object item. See also... Figure 1 The Redis database 400 stores various parameters used by the recommendation system when recalling products. Before retrieving co-occurring products of a target product, the parameters in the Redis database 400 need to be updated based on the data (user, product) obtained in step S3a. (See reference...) Figure 3 The specific explanation is as follows:

[0026] Step S1b: Query the list of products clicked by the user, and select the target product (Item). 目 Add the product to the list of products clicked by the target user. As shown in Table 1, the Redis database 400 stores the list of products clicked by platform users, which records the products clicked by each user within a preset time window (e.g., the last 3 days). After obtaining the data content (user, product) in step S3a, the product is added to the list of products clicked by the target user in Table 1, as shown in the last row of Table 1, User 目 For the target user in this example, add an object product (Item) to the list of products clicked by that user. 目 ).

[0027] Table 1: List of Products Clicked by Users

[0028] User ID List of clicked products (id) User 1 Item 1, Item 2, ... User 2 Item 3, Item 5, ... User 3 Item 1, Item 6, ... User 4 Item 2, Item 9, ... User 5 Item 4, Item 6, ... …… …… <![CDATA[User 目 ]]> <![CDATA[Item 7,Item 90,……Item 目 ]]>

[0029] Step S2b: Update the product click statistics table based on the target product. The Redis database also stores a product click statistics table, as shown in Table 2, which records the number of clicks each product received within a preset time window (e.g., the last 3 days). After obtaining the data content (user, product) in step S3a, the target product Item... 目 The number of clicks is incremented by 1, as shown in the last row of Table 2.

[0030] Table 2: Product Click Count Statistics

[0031] Step S3b involves obtaining multiple co-occurring product pairs (items) with the target product based on the target user's product click list. The Redis database 400 stores a product co-occurrence table, as shown in Table 3. The product co-occurrence table in Table 3 is obtained by counting the number of times two products clicked by the same user in Table 1. In this embodiment, co-occurring product pairs (Items) jointly clicked by the target user are added to Table 1. 目 ,Item7), (Item 目 , Item 90).

[0032] Table 3: Co-occurrence Table of Goods

[0033] Product ID Co-occurring Product (ID): Co-occurrence frequency (n) Item 1 Item 2 5 Item 1 Item 4 6 Item 1 Item 3 2 …… …… …… Item 2 Item 9 4 …… …… …… Item 3 Item 5 6 …… …… …… <![CDATA[Item 目 ]]> Item 7 7+1＝8 …… Item 90 1 …… …… ……

[0034] Step S4b: Check if the co-occurrence table of items in Table 3 includes the two newly added co-occurrence item pairs. If Table 3 already stores these two co-occurrence item pairs (Item... 目 ,Item7), (Item 目 If Item 90 is found, its co-occurrence count is incremented by 1 in step S5b. If not, the co-occurring item pair is added to Table 3 in step S6b. As shown in the last two rows of Table 3, co-occurring item pairs (Item 90) already exist in Table 3. 目 Item 7, whose co-occurrence count was originally 7, has increased to 8 after adding co-occurrence pairs obtained based on target user behavior. Table 3 originally did not contain any co-occurrence item pairs (Item 7). 目 Item90), based on target user behavior, will include co-occurring product pairs (Item... 目 Item90) is added to Table 3 and its co-occurrence count is set to 1.

[0035] In addition, to maintain the timeliness of the Redis database parameters, the parameters need to be updated according to time. Therefore, the product list in Table 1 also includes the product click time, as shown in Table 4. By monitoring the product click times in Table 4, products whose click times exceed a preset time window can be deleted. For example, if the current time is 10:16 AM on September 7, 2021, and the time window length is set to 3 days in this embodiment, products before 10:16 AM on September 4, 2021, exceed this time window, such as Item 2 in Table 4, and are therefore deleted from Table 4. Simultaneously, co-occurring product pairs, including Item 2 clicked at 23:08:28 on September 3, 2021, are deleted from the co-occurrence table. To maintain each co-occurring product pair as a pair of products clicked within a preset time window, the co-occurrence table also includes the furthest click time in each product pair, as shown in Table 5. As can be seen from the table, the earliest click time for the first co-occurrence of the co-occurring product pair (Item1, Item2) consisting of Item 2 is 2021-09-03-23:08:28, which corresponds to the record in Table 4. Therefore, it is deleted, and the number of co-occurrences is reduced by 1, from 5 times to 4 times.

[0036] Table 4: List of Products Clicked by Users

[0037]

[0038]

[0039] Table 5. Commodity Co-occurrence Table

[0040]

[0041] Step S5a: Calculate the similarity between the target product and each co-occurring product. There are multiple co-occurring product pairs including the target product, as shown in Table 3. The similarity of each co-occurring product pair is calculated separately. Due to the special nature of secondhand goods, this invention uses the number of clicks and the number of co-occurrences to measure the similarity between two products. In one embodiment, the similarity between two co-occurring products is calculated using Formula 1-1.

[0042]

[0043] Where, δ (i,j) Let n be the similarity between the i-th object item and the j-th co-occurring item; (i,j) N represents the number of times the i-th item and the j-th co-occurring item appear together; i N represents the number of clicks on the i-th item; j Let be the number of clicks on the j-th co-occurring item.

[0044] In this embodiment, it is assumed that step S3a currently yields (User1, Item1) as (User, Item), and there are 25 item pairs that co-occur with item1. Therefore, the similarity values ​​δ of the 25 items are obtained using formula 1-1. (1,1) δ (1,2) 、……δ (1,25) .

[0045] Step S6a: Sort the items in descending order of similarity, and group the top n co-occurring items into a set of similar items. The value of n is determined based on the actual situation, for example, it can be a fixed value such as 5 or 10, or it can be determined according to the needs of the recommendation process. For example, if a recommendation page has been determined, the value of n can be determined based on the number of items that can be arranged on the page, or it can be determined according to the proportion of the total number of co-occurring items.

[0046] Step S7a: Determine whether the recommendation conditions are met. If they are met, then in step S8a, recommend products from the similar product set to the target user. If they are not met, then in step S9a, monitor the target user's behavior on the platform and return to step S7a.

[0047] Figure 4 This is a block diagram illustrating the principle of a recommendation system in a second-hand e-commerce scenario according to an embodiment of the present invention. Figure 5 This is a block diagram illustrating the principle of a product recall system in a second-hand e-commerce scenario according to an embodiment of the present invention. See also... Figure 4 The recommendation system 200 includes a monitoring module 201, a product recall system 202, and a recommendation module 203. (The reference is missing from the original text.) Figure 1 The monitoring module 201 monitors user behaviors such as logging into the second-hand e-commerce platform 100 (e.g., a website or app), entering the platform's main page, and browsing products in the product list. In one embodiment, when the monitoring module 201 detects user activity on the second-hand e-commerce platform 100, it sends a trigger message to the product recall system 202. Upon receiving the trigger message, the product recall system 202 recalls similar products to the user's clicked product to obtain a set of similar products stored in the Redis database 400. The recommendation module 203 is connected to the monitoring module 201. When it determines that products need to be recommended, it retrieves similar products of the target user from the Redis database 400 and recommends them to the target user.

[0048] See Figure 5The product recall system 202 includes a data acquisition module 1, a data parsing module 2, a co-occurring product acquisition module 3, a similarity calculation module 4, and a recall module 5. After receiving a trigger message from the monitoring module 201, the data acquisition module 1 obtains real-time user behavior data from a data source 6, such as... Figure 1 The recommendation system 200 connects to Kafka data source 300 and deploys a Spark-Streaming real-time task to it. Data acquisition module 1 extracts real-time behavior messages for the target user from the Kafka data source via the consumer API. Data parsing module 2, connected to data acquisition module 1, extracts the user's identity identifier and the object product identity identifier of the user's current behavior based on the user's real-time behavior data, thus obtaining a data entry containing (user, product). Co-occurring product acquisition module 3, connected to data parsing module 2, retrieves data from parameter database 7 (e.g., ...). Figure 1 The similarity calculation module 4 is connected to the co-occurring product acquisition module 3, and calculates the similarity between the object product and each co-occurring product. The module retrieves co-occurring products from the Redis database 400.

[0049] In one embodiment, such as Figure 6 As shown, the product recall system 202 further includes a parameter determination module 8, which comprises a parameter data acquisition unit 81, a co-occurrence product pair determination unit 82, and a statistics unit 83. The parameter data acquisition unit 81 acquires multiple historically clicked products from a recent preset time window, as shown in Tables 1 and 4, and stores them in the parameter database 7. The co-occurrence product pair determination unit 82 is connected to the parameter data acquisition unit 81 and identifies two products clicked by the same user as a co-occurrence product pair, as shown in Tables 3 and 5. The statistics unit 83 is connected to both the parameter data acquisition unit 81 and the co-occurrence product pair determination unit 82, and counts the number of clicks on each product and the number of co-occurrences of each co-occurrence product pair, obtaining the content shown in Tables 1-5, and storing it in the parameter database 7.

[0050] like Figure 7 As shown, the similarity calculation module 4 includes a query unit 41 and a calculation unit 42. The query unit 41 queries the parameter database 7 for the number of clicks on the object product, the number of clicks on the co-occurring products of the object product, and the number of co-occurrences of the co-occurring product pairs formed by the object product. The calculation unit 42 calculates the similarity between the object product and its co-occurring products according to formula 1-1:

[0051]

[0052] Where, δ (i,j) Let n be the similarity between the i-th object item and the j-th co-occurring item; (i,j) N represents the number of times the i-th item and the j-th co-occurring item appear together; i N represents the number of clicks on the i-th item; j Let j be the number of clicks on the j-th co-occurring product. The calculation unit 42 calculates the similarity value of multiple co-occurring product pairs using formula 1-1 and stores it in the parameter database 7. The recall module 5 is connected to the similarity calculation module 4. It sorts the obtained similarity values ​​of multiple co-occurring product pairs in descending order of similarity and selects a preset number of co-occurring products at the top of the list as recall products, storing them in the similar product set of the target user. Figures 5-7 The parameter database 7 in the middle can be Figure 1 The Redis database in the middle is 400.

[0053] Among them, the parameter data in parameter database 7, such as the user-clicked product list, product click table, and co-occurrence table, are several crucial parameters in the product recall process. To ensure that the recalled products match user interests, it is necessary to maintain the timeliness of this data in real-time or periodically, promptly delete data exceeding the time window, and add clicked products to parameter database 7 in a timely manner based on user click behavior on the platform. Therefore, in Figure 8 In the illustrated embodiment, the product recall system 202 further includes an update module, which mainly comprises a first update trigger module 91 and a second update trigger module 92. The first update trigger module 91 is connected to the data parsing module 2. When the data parsing module 2 parses the user's product click behavior and obtains the data (user, product), it sends a first update instruction to the parameter determination module 8. (See also...) Figure 7The parameter data acquisition unit 81 in the parameter determination module 8 adds the product identifier from the first update instruction to the list of products clicked by the user, and notifies the co-occurrence product pair determination unit 82 and the statistics unit 83. The co-occurrence product pair determination unit 82 obtains new co-occurrence product pairs based on the newly added object product. The statistics unit 83 queries the co-occurrence product table to see if there are any new co-occurrence product pairs. If so, its co-occurrence count is incremented by 1; otherwise, the new co-occurrence product pair is added to the co-occurrence product table. The second update triggering module 92 monitors the current time and issues a second update instruction to the parameter determination module 8 when the current time window length exceeds the preset time window length. The parameter data acquisition unit 81 in the parameter determination module 8 deletes user-hidden clicked products that exceed the preset time window based on the time information. Correspondingly, the co-occurrence product pair determination unit 82 deletes the co-occurrence product pairs formed with the deleted products in the co-occurrence product table based on the deleted user-hidden clicked products, thereby modifying the co-occurrence count and decrementing the current co-occurrence count by 1. The statistics unit 83 modifies the click count of the deleted product based on the deleted product, decrementing the current count by 1. This ensures that the data stored in parameter database 7 for calculating product similarity is always up-to-date within the latest time window, thus guaranteeing that similar products are provided based on the user's latest interests.

[0054] The above embodiments are for illustrative purposes only and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the scope of the invention. Therefore, all equivalent technical solutions should also fall within the scope of the invention.

Claims

1. A method for recalling recommended products in a second-hand e-commerce scenario, comprising: Obtain real-time behavioral data of the target user; Extract the user identity identifier of the target user and the object product identity identifier of the target user's current real-time behavior from the user's real-time behavior data; Retrieve co-occurring products that form co-occurring product pairs with the object product from the parameter database; Calculate the similarity between the object product and each co-occurring product; and Multiple co-occurring products are sorted in descending order of similarity, and the top-ranked co-occurring products are selected as the recalled products. The similarity between the target product and its co-occurring products is calculated according to Formula 1-1: 1-1 in, Let be the similarity between the i-th object item and the j-th co-occurring item; This represents the number of times the i-th item and the j-th co-occurring item co-occur. This represents the number of clicks on the i-th item. Let be the number of clicks on the j-th co-occurring item.

2. The method of claim 1, further comprising: Real-time behavioral messages are extracted from a Kafka data source by deploying spark-streaming real-time tasks to the Kafka data source. The message is parsed to obtain the user's identity and behavior type; as well as If the user behavior type is a product click, obtain the product identity identifier of the clicked object.

3. The method of claim 1, further comprising the step of obtaining parameters: Retrieve the historical clicked products of multiple users within a recent preset time window; Count the number of clicks for each product; and Count the number of times co-occurring product pairs clicked by the same user.

4. The method of claim 3, further comprising: Update the user's historical clicks on the product, the number of clicks on the target product, and the number of times the product pairs co-occurred with the target product based on the user's real-time click behavior.

5. The method of claim 4, further comprising: Delete historical clicked items by users that have exceeded a preset time window based on time information; as well as Update the number of times the product was clicked and the number of times its co-occurring product pairs were co-occurred based on the deleted user history clicks.

6. A recommended product recall system for a second-hand e-commerce scenario, comprising: The data acquisition module is configured to acquire real-time behavioral data of the target user; The data parsing module, which is connected to the data acquisition module, is configured to extract the user identity identifier of the target user and the object product identity identifier of the target user's current behavior from the target user's real-time behavior data; A co-occurring product acquisition module, which is connected to the data parsing module, is configured to acquire co-occurring products that form co-occurring product pairs with the object product from a parameter data source; A similarity calculation module, which is connected to the co-occurring product acquisition module, is configured to calculate the similarity between the object product and each co-occurring product. as well as The recall module is connected to the similarity calculation module and is configured to sort multiple co-occurring products in descending order of similarity, and to use the first preset number of co-occurring products as the recall products. The similarity calculation module includes: The calculation unit calculates the similarity between the target product and its co-occurring products according to Formula 1-1: 1-1 in, Let be the similarity between the i-th object item and the j-th co-occurring item; This represents the number of times the i-th item and the j-th co-occurring item co-occur. This represents the number of clicks on the i-th item. Let be the number of clicks on the j-th co-occurring item.

7. The system according to claim 6, further comprising a Kafka data source; the data acquisition module extracts real-time behavioral messages from the Kafka data source by deploying a Spark-Streaming real-time task to the Kafka data source.

8. The system according to claim 6, further comprising a parameter determination module and a parameter database, wherein, The parameter determination module includes: The parameter data acquisition unit is configured to acquire multiple historical clicked products by users within a recent preset time window; A co-occurrence product pair determination unit, connected to the parameter data acquisition unit, is configured to determine two products clicked by the same user as a co-occurrence product pair; and A statistics unit, connected to the parameter data acquisition unit and the co-occurrence product pair determination unit, is configured to count the number of clicks on each product and the number of co-occurrences of each co-occurrence product pair; The parameter database is connected to the parameter determination module and is configured to store the user's historical clicked products, the number of times the products were clicked, and the number of times co-occurring co-occurring product pairs within the most recent preset time window.

9. The system according to claim 8, wherein the similarity calculation module comprises: The query unit is configured to query the parameter database based on the object product for the number of times the object product has been clicked, the number of times the object product's co-occurring products have been clicked, and the number of times the object product's co-occurring product pairs have co-occurred.

10. The system according to claim 8, further comprising a first update triggering module connected to the data parsing module and the parameter determination module, wherein the first update triggering module is configured to send a first update instruction to the parameter determination module when the data parsing module parses that the user clicks on a product in real time, and the parameter determination module updates the user's historical clicked products, the number of clicks on the target product, and the number of co-occurrences of product pairs co-occurring with the target product according to the target product in the update instruction.

11. The system according to claim 10, further comprising a second update triggering module connected to the parameter determining module, configured to monitor the current time, and issue a second update instruction to the parameter determining module when the length of the current time window exceeds the length of a preset time window, wherein the parameter determining module deletes user-hidden clicked items that exceed the preset time window based on time information, and correspondingly updates the number of clicks of the item and the number of co-occurrences of its co-occurring item pairs based on the deleted user-hidden clicked items.

Images