WO2020001096A1 - 商品推荐方法和商品推荐设备 - Google Patents

商品推荐方法和商品推荐设备 Download PDF

Info

Publication number
WO2020001096A1
WO2020001096A1 PCT/CN2019/079433 CN2019079433W WO2020001096A1 WO 2020001096 A1 WO2020001096 A1 WO 2020001096A1 CN 2019079433 W CN2019079433 W CN 2019079433W WO 2020001096 A1 WO2020001096 A1 WO 2020001096A1
Authority
WO
WIPO (PCT)
Prior art keywords
user
recommended
sample
product
users
Prior art date
Application number
PCT/CN2019/079433
Other languages
English (en)
French (fr)
Inventor
李慧
Original Assignee
京东方科技集团股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 京东方科技集团股份有限公司 filed Critical 京东方科技集团股份有限公司
Priority to US16/500,656 priority Critical patent/US11449919B2/en
Publication of WO2020001096A1 publication Critical patent/WO2020001096A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00Commerce
    • G06Q30/06Buying, selling or leasing transactions
    • G06Q30/0601Electronic shopping [e-shopping]
    • G06Q30/0631Item recommendations
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00Machine learning
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00Computing arrangements using knowledge-based models
    • G06N5/04Inference or reasoning models
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00Commerce
    • G06Q30/02Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201Market modelling; Market analysis; Collecting market data
    • G06Q30/0204Market segmentation
    • G06Q30/0205Location or geographical consideration
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00Commerce
    • G06Q30/06Buying, selling or leasing transactions
    • G06Q30/0601Electronic shopping [e-shopping]
    • G06Q30/0639Item locations

Definitions

  • the present disclosure relates to the field of communication technologies, and in particular, to a method and a device for recommending products.
  • Product recommendation can recommend the right product to the user in the right scene.
  • e-commerce recommendation systems that recommend various products to users on the Internet, such as recommending new shelves, discounted products, and popular products.
  • the offline retail industry is also changing. Offline retailers need to rely on new technologies to develop new shopping models. At present, there is a lack of a product recommendation method for users' offline shopping behavior.
  • a product recommendation device including:
  • An information acquisition module configured to receive location information of a user to be recommended
  • a product recommendation module configured to perform product recommendation according to the location information of the user to be recommended
  • the information sending module is configured to send the recommended product information to a user to be recommended.
  • the product recommendation module includes:
  • a hotspot area acquisition module configured to obtain a hotspot area
  • a location judgment module configured to determine whether the user to be pushed is in a hotspot area according to the location information of the user to be recommended;
  • the first product recommendation sub-module is configured to perform product recommendation based on the historical shopping behavior of multiple first sample users when the user to be recommended is not in the hot area;
  • a second product recommendation module configured to perform product recommendation based on the historical shopping behavior of the plurality of first sample users and the sales volume of products in the hotspot when the user to be recommended is in the hotspot;
  • the plurality of first sample users include users to be recommended.
  • the hotspot area acquisition module includes:
  • a position acquisition module configured to acquire position coordinates of a plurality of second sample users
  • a clustering module configured to use a clustering algorithm to cluster the plurality of second sample users based on the position coordinates of the plurality of second sample users to obtain at least one second sample user class
  • the hotspot area determination module is configured to determine a hotspot area according to the at least one second sample user class, where a center point coordinate of the hotspot area corresponding to each second sample user class And radius r are determined by:
  • n is the number of users in the second sample user class corresponding to the hotspot area
  • the first submodule of the product recommendation includes:
  • a data acquisition module configured to acquire historical shopping behavior data of the plurality of first sample users
  • a scoring matrix determining module configured to determine a scoring matrix of a first sample user-item according to the historical shopping behavior data
  • a predictive scoring matrix determination module configured to determine a predictive scoring matrix for the first sample of user-commodities based on a scoring matrix of the first sample of users-commodities by using a collaborative filtering algorithm based on matrix decomposition;
  • the user terminal is configured to send the location information of the user to be recommended to the product recommendation device;
  • the product recommendation device is configured to receive location information of the user to be recommended, perform product recommendation based on the location information of the user to be recommended, and send the recommended product information to the user to be recommended;
  • product recommendation is performed based on the historical shopping behavior of multiple first sample users, where the multiple first sample users include users to be recommended;
  • product recommendation is performed based on the historical shopping behavior of the plurality of first sample users and the sales volume of the goods in the hotspot area.
  • the step of obtaining a hotspot area includes:
  • the plurality of first sample users are m first sample users U1, ..., Um, and historical shopping of the plurality of first sample users
  • the behavior involves n kinds of products V1, ..., Vn, and the step of determining a scoring matrix of the first sample user-item according to the historical shopping behavior data includes:
  • Vj the score of each of the first sample users Ui on the products V1, ..., Vn, where Ui is the case where the historical shopping behavior of the first sample user Ui does not involve the product Vj.
  • the predictive score of the first sample user-item is determined by using a matrix filtering-based collaborative filtering algorithm based on the first sample user-item scoring matrix.
  • the steps of the matrix include:
  • a ′ U ⁇ V T ,
  • the element A ′ (i, j) of the matrix A ′ represents the prediction score of the product Vj by the user Ui.
  • the step of performing product recommendation to the user to be recommended according to the first sample user-product prediction score matrix includes:
  • the step of determining whether the user to be recommended is in a hot area according to the location information of the user to be recommended includes:
  • the step of performing product recommendation based on the historical shopping behavior of the plurality of first sample users and the sales volume of the products in the hotspot area when the user is in the hotspot area includes:
  • N is an integer greater than or equal to 1,
  • Product recommendation is performed according to the first sample user-product prediction score matrix and the N products ranked first in the hotspot area.
  • the step of performing product recommendation according to the first sample user-item prediction score matrix and the N products ranked first in the hotspot area includes:
  • the first and second recommended products are recommended to users to be recommended in a fixed or random priority order.
  • the collaborative filtering algorithm based on matrix factorization includes a collaborative filtering algorithm based on matrix decomposition of a cross-least square method or a collaborative filtering algorithm based on matrix decomposition of a gradient descent method.
  • the method further includes:
  • the product that has been put in the shopping cart is deleted from the recommended product information.
  • the step of obtaining product information of the user to be recommended that has been placed in a shopping cart includes: using a product identification device on the shopping cart to obtain Product information.
  • a computing device including: a processor; and a memory having computer-readable instructions stored thereon, which when executed by the processor, causes the computing device to execute Product recommendation method according to some embodiments of the present disclosure.
  • a computer-readable storage medium including computer-readable instructions stored thereon, which computer-readable instructions, when executed, implement a method for recommending goods according to some embodiments of the present disclosure.
  • FIG. 1 shows a flowchart of a product recommendation method according to some embodiments of the present disclosure
  • FIG. 3 shows a flowchart of obtaining a hotspot area in a method for recommending products according to some embodiments of the present disclosure
  • FIG. 6 shows a flowchart of a method for recommending products according to other embodiments of the present disclosure
  • FIG. 7 schematically illustrates a structural block diagram of a product recommendation device according to some embodiments of the present disclosure.
  • FIG. 1 shows a flowchart of a product recommendation method according to some embodiments of the present disclosure.
  • a product recommendation device may be provided for performing the product recommendation method shown in FIG. 1.
  • the product recommendation device may be any computing device with data transmission and processing capabilities, such as a server, a computer, and the like.
  • a product recommendation method according to some embodiments of the present disclosure includes the following steps S101-S103.
  • the above position information may be position coordinates of a user when shopping in a mall or a supermarket.
  • the position coordinates may be expressed by using coordinate points in a plane (two-dimensional) or space (three-dimensional) rectangular coordinate system.
  • the mall contains only one
  • a plane coordinate system can be used, and if the shopping mall is large and contains multiple layers, a spatial coordinate system is required to uniquely represent each location in the shopping mall.
  • the position coordinates can also be expressed in other ways, such as polar coordinates and the like. In the following, a mall with only one floor will be used as an example.
  • a positioning function such as a GPS module
  • a user terminal such as a mobile phone or an electronic device installed on a shopping cart
  • location information of the user terminal can be regarded as the location information of the user.
  • the user terminal may send location information to the server at a preset time.
  • the product recommendation device or other server in the shopping mall stores user records or member information corresponding to the user identity information or user ID.
  • user ID Generally, when a user makes a purchase in a mall for the first time, he can register as a member and enter membership information; after registration, the user will get a membership card. Since each membership card corresponds to a unique card number, the membership card number can be used.
  • the user ID As the user ID; subsequently, the user ID is entered into the server, or the user's registration information is directly entered into the server and the user ID is generated by the server and returned to the user for use by the user when shopping in the mall.
  • the user ID may be represented by, for example, a one-dimensional code, a two-dimensional code, or a card number.
  • the product recommendation device can recommend products near the location to the user, such as promotional products or best-selling products, according to the user's location. In addition, you can also recommend products near the location based on the user ’s historical shopping history.
  • product recommendation may be performed in different ways according to different locations of the user.
  • a linear shopping place such as a mall or a supermarket
  • the mall can be divided into hotspot areas (that is, hot sales areas or crowded areas) and non-hotspot areas, and then use different algorithms to recommend products based on whether the user is in the hotspot area, that is, in non-
  • the hotspot area can adopt the collaborative filtering algorithm, while the hotspot area uses the collaborative filtering algorithm combined with the topN algorithm of product sales.
  • This location-based recommendation method can better meet the needs of users, and further improve the accuracy of product recommendation and the convenience of shopping.
  • the recommended product information can be sent to the user terminal through a wired or wireless network (such as wifi, etc.), so as to inform the user of the related information of the recommended product, such as the product name, specific location in the mall, etc. For your reference.
  • a wired or wireless network such as wifi, etc.
  • the shopping area of the mall or supermarket may be divided or classified into several different areas in advance, and then different product recommendation methods may be adopted according to different areas where the user is located.
  • the hotspot area described in step S1021 represents a crowded area in a shopping mall or an area where shoppers stay longer, and may be a predetermined area.
  • the hotspot area may be set in advance according to the layout of the goods in the shopping mall. ; Or it can be set by the user ’s location information or movement track data within a period of time, for example, by clustering users with small changes in location within a certain period of time according to their location coordinates to determine the hotspot area, that is, where such users are located
  • the area is the hotspot area. For details, see the description below in conjunction with FIG. 3.
  • step S1023 The purpose of step S1023 is to predict the shopping behavior of the current user according to the shopping behavior of the target user or to-be-recommended users and other users in the mall, thereby recommending products to the target user.
  • This recommendation method is based on the following ideas. Different users have a certain degree of similarity in the purchase of goods, so it is a reference to combine the purchasing habits of all users.
  • the first sample users may include all users registered as members in the mall.
  • the first sample user may also include only some member users, excluding those member users who have no shopping records or have fewer shopping records, in order to simplify the calculation amount and improve the accuracy of the recommendation.
  • the first sample of users can be updated regularly to keep pace with the times and more in line with the actual situation of users' shopping behavior.
  • User shopping behavior refers to the behavior or action related to the purchase of a product when the user is shopping in a mall or a supermarket, such as browsing a product, viewing a product, placing a product in a shopping cart, and the like.
  • the user's shopping behavior can be roughly divided into four types: no product viewed, only product viewed, product viewed and placed in a shopping cart but ultimately not purchased, and purchased.
  • the above division method is only used as an example. In actual applications, other methods may also be used to identify a user's shopping behavior.
  • the above-mentioned shopping behavior identifiers can be converted into numbers, and these numbers can be regarded as the user's scores on the corresponding products, so that multiple first sample users U1-Um can be scored on multiple products V1-Vn .
  • a matrix filtering-based collaborative filtering algorithm can be used to predict the user's unrated (that is, the user's shopping behavior is "null") products. And recommend the product with the highest predicted score to the user.
  • the historical shopping behavior based on multiple first sample users (including the user to be recommended) in step S1023 i.e., all Or part of the user recommends products in the shopping mall in the manner of shopping or rating) to recommend products to the user more accurately and accurately.
  • FIG. 3 shows a flowchart of obtaining a hotspot area in a method for product recommendation according to some embodiments of the present disclosure.
  • a hotspot area refers to an area where people are crowded in a mall or an area where shoppers stay longer. Therefore, in addition to the above-mentioned pre-determined hotspots based on the product layout or historical shopping information, hotspots can also be determined based on the location of all customers or consumers currently shopping in the mall, such as customer-intensive areas, That is, the area where the location points are concentrated can be defined as a hotspot area. Therefore, customers whose location points are concentrated in a certain area can be classified into one category, and the area where such customers are located can be determined as a hot area.
  • the position coordinates of all users in the current shopping mall can be used as sample points, and these sample points (or corresponding users) are clustered by a clustering algorithm to obtain one or more classes of user (location points) (i.e. A set of users whose location points are close to each other), and the area where the one or more categories of users are located is a hotspot area.
  • location points i.e. A set of users whose location points are close to each other
  • step S1021 of acquiring a hotspot area shown in FIG. 2 includes the following steps S1021a-S1021c.
  • S1021a Obtain position coordinates of a plurality of second sample users.
  • the second sample user may refer to all customers currently in the shopping area of the mall, where the shopping area is an area in the mall excluding the mall entrance, exit, and elevator areas.
  • the second sample user may also be a part of customers who are shopping in the mall, for example, it may be limited to users who have a small position change within a certain period of time (that is, customers who stay in a certain area for a long time), In this way, the area where it stays is closer to the hotspot area.
  • the second sample user may be determined according to the movement trajectory of the customer.
  • the second sample user can also locate himself through a positioning function integrated in a user terminal (such as a mobile phone or an electronic device installed on a shopping cart), and use the terminal to position information Product recommendation equipment sent to the mall or supermarket.
  • a user terminal such as a mobile phone or an electronic device installed on a shopping cart
  • the above position information may be position coordinates of a user when shopping in a mall or a supermarket, and the position coordinates may be represented by coordinates (x, y) in a plane rectangular coordinate system, for example.
  • Input data sample D, where data sample D is the coordinate point of the second sample user;
  • Step 1 Establish two queues, which are an ordered queue (data to be processed) and a result queue (processed data).
  • the ordered queue is used to store the core object and its directly reachable objects. The distances are arranged in ascending order; the result queue is used to store the output order of the sample points;
  • Step 3 if the ordered sequence is empty, go back to step 2, otherwise take the first point from the ordered queue;
  • step 3 Repeat step 3 until the ordered queue is empty;
  • Step 4 Take the points from the result queue in order. If the reachable distance of the point is not greater than the given radius ⁇ , the point belongs to the current category, otherwise go to step 5;
  • Step 5 If the core distance of the point is greater than a given radius ⁇ , the point is noise and can be ignored, otherwise the point belongs to a new class, skip to step 1;
  • Step 6 When the result queue traversal ends, the algorithm ends.
  • n is the number of users in the second sample user class corresponding to the hotspot area
  • (x i , y i ) are position coordinates of the i-th user in the second sample user class corresponding to the hotspot area.
  • the obtained hotspot area may be a circular area. Therefore, when determining whether the user to be recommended is in a hotspot area, the position of the user to be recommended and the center point of the hotspot area may be calculated. The Euclidean distance between them is compared with the radius r of the hotspot area to achieve: if d> r, the user is not in the hotspot area; if d ⁇ r, the user is in the hotspot area.
  • hotspots of different shapes may also be defined in other ways.
  • FIG. 4 shows a flowchart of product recommendation based on a historical shopping behavior of a first sample user in a product recommendation method according to some embodiments of the present disclosure.
  • step S1023 shown in FIG. 2 further includes the following steps S1023a-S1023d.
  • S1023a Obtain historical shopping behavior data of multiple first sample users.
  • the historical shopping behavior data of the first sample user may be set in the following manner:
  • the first sample user When the first sample user does not view the product, the first sample user identifies the shopping behavior data of the product as a null value;
  • the first sample user When the first sample user only views the product, the first sample user identifies the shopping behavior data of the product as "view";
  • the first sample user When the first sample user views the product and puts the product in the shopping cart but does not purchase it, the first sample user identifies the shopping behavior data of the product as "put into the shopping cart";
  • the first sample user When a first sample user purchases a product, the first sample user identifies the shopping behavior data of the product as "buy”.
  • S1023b Determine a first sample user-item scoring matrix according to the historical shopping behavior data.
  • Table 2 is an exemplary first sample user-item scoring table, where U1-U5 respectively represent different second sample users, V1-V4 respectively represent different products, and the values in the table are the second sample user pairs.
  • Table 2 The first sample user-item score table
  • the above scoring table can be converted into the form of a scoring matrix. Therefore, for the m first sample users U1, ..., Um and n kinds of products V1, ..., Vn, according to the historical shopping behavior data of the first sample users, as shown in Table 1 above. Scoring strategy to obtain an m ⁇ n first sample user-item rating matrix A.
  • a matrix of elements A (i, j) indicates that the user Ui commodity Vj rating, the rating may represent a level of interest the user Ui Vj, commodities, with higher scores indicating that the user U i bought the goods Vj.
  • the score is "null” or missing (represented by "-"), it means that the user Ui's shopping behavior for the product Vj is "not viewed", that is, there is no record of Vj in the historical shopping behavior record of the first sample user Ui. Historical shopping behavior (and therefore cannot be scored). Therefore, these missing values in the matrix A are the predicted scores of the user Ui for the product Vj that we want to determine, and the process of the predicted scores is the null value completion process of the matrix A.
  • a matrix without missing values obtained by completing the null or missing completion of the rating matrix A m ⁇ n of the first sample user-product may be referred to as the first The user-item prediction score matrix A ′.
  • a collaborative filtering algorithm based on matrix decomposition of alternating least squares (ALS) can be used.
  • the m ⁇ n scoring matrix A can use two small matrices U m ⁇ k and V n ⁇ The product of k is approximated by: A ⁇ U ⁇ V T , k ⁇ m, n.
  • the matrices U and V can be obtained, so that the product of U and V can be used to restore the user-product scoring matrix.
  • the prediction score matrix A ′ U ⁇ V T of the first sample user-item is obtained.
  • the calculation engine Spark can be used for model training and tuning.
  • a matrix decomposition collaborative filtering algorithm based on a gradient descent method may also be used to determine the prediction sample matrix of the first sample user-product.
  • S1023d Perform product recommendation to the user to be recommended according to the prediction sample matrix of the first sample user-product.
  • the ratings of all the first sample users for all the products in the mall can be determined, and since the first sample user includes users to be recommended, the prediction score matrix can be obtained from In the prediction scores of users to be recommended for all products in the mall. Therefore, products can be recommended to users based on the level of the predicted score. For example, the user's prediction score for each product is arranged in order from high to low, and the first M products are selected and sorted, where M is a positive integer.
  • the recommended product is determined according to the predicted score of the user on various products, so that the recommendation scheme more accurately meets the needs and wishes of the user, and enhances For a personalized experience.
  • FIG. 5 shows a flowchart of product recommendation based on historical shopping behaviors of multiple first sample users and sales of products in hotspots in a method of product recommendation according to some embodiments of the present disclosure.
  • step S1024 in FIG. 3 includes the following steps:
  • S1024a sort the products in the hotspot area according to the sales volume from high to low to obtain the N products that are ranked first, where N is an integer greater than or equal to 1.
  • S1024e Perform product recommendation based on the first sample user-item prediction scoring matrix and the first N products ranked in the hotspot area.
  • a combination of two methods is used to determine a recommended product.
  • the first method is to make product recommendations based on the sales volume of products located in hot areas (that is, the TopN algorithm); the second method is to use the first sample users (for example, all users or members with shopping behavior records in the mall) )
  • To make product recommendations in the historical shopping behavior which is the same way as when the user is not in the hot area, see the flowchart shown in FIG. 4.
  • the above combination method considers both the characteristics of the user (historical shopping habits of himself and other users) and the characteristics of the product (sales and location of the product), so as to better meet the needs of the user and further improve the product recommendation Accuracy and convenience of shopping.
  • a number of recommended products can be obtained by using a collaborative filtering algorithm and a topN algorithm, respectively, and then the obtained different recommended products can be recommended to users in different priorities. These recommended products are fixed or randomly placed in different recommendation positions (that is, setting recommendation priorities) for users to make their own choices.
  • a number of recommended products can be obtained by using a collaborative filtering algorithm and a topN algorithm, respectively, and then the obtained different recommended products can be recommended to users in different priorities.
  • These recommended products are fixed or randomly placed in different recommendation positions (that is, setting recommendation priorities) for users to make their own choices.
  • it can be set randomly or fixed first, and then dynamically adjusted according to the specific situation of the user's selection (click). For example, if the click-through rate of the recommended products obtained by the topN algorithm is high, such products are placed in the top recommendation position for priority recommendation; otherwise, the recommended products obtained by the collaborative filtering algorithm are given priority recommendation.
  • step S1024e shown in FIG. 5-according to the first sample user-the prediction score matrix of the product and the N products ranked first in the hotspot area may include:
  • the first and second recommended products are recommended to users to be recommended in a fixed or random priority order.
  • FIG. 6 shows a flowchart of a product recommendation method according to other embodiments of the present disclosure. Compared with FIG. 1, the product recommendation method described in FIG. 6 further includes the following steps S104-S105 after step S102 and before step S103.
  • a camera or a scanning device is set on the shopping cart.
  • a camera or a scanning device on the shopping cart obtains product information of the product.
  • the product information may include a product ID, and may also include a product name, a category to which the product belongs, and a product price.
  • the product information obtained by the shopping cart will be further sent to the product recommendation device.
  • the recommended product information includes a product that has been placed in the shopping cart, delete the product that has been placed in the shopping cart from the recommended product information.
  • the server After the server determines the recommended product, it can query the recommended product based on the obtained product information. If the recommended product information includes the product corresponding to the product information (hereinafter referred to as B product), it means that the user has included the recommended product at this time. B product in the shopping cart, in order to avoid the user repeatedly buying A product with reference to the recommended product, the B product included in the recommended product information may be deleted to obtain new recommended product information.
  • B product the product included in the recommended product information may be deleted to obtain new recommended product information.
  • FIG. 7 schematically illustrates a structural block diagram of a product recommendation device according to some embodiments of the present disclosure.
  • the product recommendation method according to some embodiments of the present invention may be completed by a product recommendation device as shown in FIG. 7.
  • the product recommendation device 700 includes:
  • the information acquisition module 710 is configured to receive location information of a user to be recommended
  • the product recommendation module 720 is configured to perform product recommendation according to the location information of the user to be recommended;
  • the information sending module 730 is configured to send the recommended product information to a user to be recommended.
  • the product recommendation module includes:
  • a location judgment module configured to determine whether the user to be pushed is in a hotspot area according to the location information of the user to be recommended;
  • the first product recommendation sub-module is configured to perform product recommendation based on the historical shopping behavior of multiple first sample users when the user to be recommended is not in the hot area;
  • a second product recommendation module configured to perform product recommendation based on the historical shopping behavior of the plurality of first sample users and sales of products in the hotspot when the user to be recommended is in the hotspot;
  • the plurality of first sample users include users to be recommended.
  • the hotspot area acquisition module includes:
  • a position acquisition module configured to acquire position coordinates of a plurality of second sample users
  • a clustering module configured to use a clustering algorithm to cluster the plurality of second sample users based on the position coordinates of the plurality of second sample users to obtain at least one second sample user class
  • the hotspot area determination module is configured to determine a hotspot area according to the at least one second sample user class, where a center point coordinate of the hotspot area corresponding to each second sample user class And radius r are determined by:
  • n is the number of users in the second sample user class corresponding to the hotspot area
  • the first submodule of the product recommendation includes:
  • a data acquisition module configured to acquire historical shopping behavior data of the plurality of first sample users
  • a predictive scoring matrix determination module configured to determine a predictive scoring matrix for the first sample of user-commodities based on a scoring matrix of the first sample of users-commodities by using a collaborative filtering algorithm based on matrix decomposition;
  • the third product recommendation module is configured to perform product recommendation to the user to be recommended according to the first sample user-item prediction score matrix.
  • a product recommendation system including a product recommendation device and a user terminal that is communicatively connected with the product recommendation device, wherein:
  • the user terminal is configured to send the location information of the user to be recommended to the product recommendation device;
  • the product recommendation device is configured to receive location information of the user to be recommended, perform product recommendation based on the location information of the user to be recommended, and send the recommended product information to the user to be recommended;
  • the user terminal is further configured to receive and display the recommended product information from the product recommendation device.
  • FIG. 8 schematically illustrates a structure diagram of a product recommendation system according to some embodiments of the present disclosure.
  • the product recommendation system includes a product recommendation device 810 and a user terminal 820 that is communicatively connected with the product recommendation device 810, where:
  • the user terminal 820 is configured to send location information of the user to be recommended to the product recommendation device;
  • the product recommendation device 810 is configured to receive position information of a user to be recommended, perform product recommendation according to the position information of the user to be recommended, and send the recommended product information to the user to be recommended;
  • the user terminal 820 is further configured to receive and display the recommended product information from the product recommendation device 810.
  • the present invention also relates to one or more computer storage media on which a computer program is stored, and when the computer program is executed, the method for recommending goods according to some embodiments of the present invention as described above is implemented.
  • the present invention also relates to a computing device, comprising: a processor; and a memory having computer-readable instructions stored thereon, which, when the computer-readable instructions are executed by the processor, cause the computing device to execute as described above A method for recommending products according to some embodiments of the present invention.
  • Any process or method description in a flowchart or otherwise described herein can be understood as representing a module, fragment, or portion of code that includes one or more executable instructions for implementing steps of a custom logic function or process, And the scope of the preferred embodiments of the present invention includes additional implementations in which the functions may be performed out of the order shown or discussed (including in a substantially simultaneous manner or in the reverse order according to the functions involved), which should be performed by Those skilled in the art to which the embodiments of the present invention pertain will understand.
  • a sequenced list of executable instructions that can be considered to implement a logical function can be embodied in any computer-readable medium,
  • the instruction execution system, device, or device such as a computer-based system, a system including a processor, or other system that can fetch and execute instructions from the instruction execution system, device, or device), or combine these instruction execution systems, devices, or devices Or equipment.
  • a "computer-readable medium” may be any device that can contain, store, communicate, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device.
  • Computer-readable media may include, for example, the following: electrical connections (electronic devices) with one or more wirings, portable computer disk cartridges (magnetic devices), random access memories (Random Access Memory), Read-only memory (Read Only Memory), erasable and editable read-only memory (Erasable, Programmable, Read Only Memory) or flash memory, optical fiber devices, and compact disc read-only memory (Compact Disc Read Only Memory).
  • the computer-readable medium may even be paper or other suitable medium on which the program can be printed, because, for example, by optically scanning the paper or other medium, followed by editing, interpretation, or other suitable Processing to obtain the program electronically and then store it in computer memory.
  • each part of the present invention may be implemented by hardware, software, firmware, or a combination thereof.
  • multiple steps or methods may be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system.
  • a suitable instruction execution system For example, if it is implemented in hardware, it may be implemented in any one of the following technologies known in the art or a combination thereof: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, a suitable combination Application-specific integrated circuits for logic gate circuits, Programmable Gate Arrays, Field Programmable Gate Arrays, etc.
  • the program may be stored in a computer-readable storage medium.
  • the program includes the execution.
  • each functional unit in each embodiment of the present invention may be integrated into one processing module, or each unit may exist separately physically, or two or more units may be integrated into one module.
  • the above integrated modules can be implemented in the form of hardware or software functional modules. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Theoretical Computer Science (AREA)
  • Strategic Management (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Development Economics (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Economics (AREA)
  • Software Systems (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Computational Linguistics (AREA)
  • Game Theory and Decision Science (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种商品推荐方法和商品推荐设备,该商品推荐方法包括接收待推荐用户的位置信息;根据待推荐用户的位置信息,进行商品推荐;以及将所推荐的商品信息发送给待推荐用户。这种基于位置的商品推荐方法能够更精确地满足用户需求,同时提高购物的便利性。

Description

商品推荐方法和商品推荐设备
相关申请
本申请要求2018年6月29日提交的申请号为201810699571.7的中国专利申请的优先权,该专利申请的所有内容通过引用合并于此。
技术领域
本公开涉及通信技术领域,尤其涉及一种商品推荐方法和商品推荐设备。
背景技术
商品推荐可以在恰当的场景给用户推荐恰当的商品。常见的如电子商务推荐系统,在互联网上给用户推荐各种商品,如推荐新的上架的商品、打折的商品和热销的商品等。目前互联网上的电子商务推荐系统种类多样。但是,随着云计算、大数据、物联网等新技术的发展,线下零售行业也在发生变化,线下零售商需要依托新技术,开发新的购物模式。目前,针对用户的线下购物行为,缺乏一种商品推荐方式。
发明内容
本公开的目的在于提供一种商品推荐方法和商品推荐设备。
根据本公开的第一方面,提供一种商品推荐设备,包括:
信息获取模块,配置成接收待推荐用户的位置信息;
商品推荐模块,配置成根据待推荐用户的位置信息,进行商品推荐;
信息发送模块,配置成将所推荐的商品信息发送给待推荐用户。
在根据本公开一些实施例的商品推荐设备中,商品推荐模块包括:
热点区域获取模块,配置成获取热点区域;
位置判断模块,配置成根据待推荐用户的位置信息,确定所述待推用户是否在热点区域中;
商品推荐第一子模块,配置成当待推荐用户不在热点区域中时,基于多个第一样本用户的历史购物行为进行商品推荐;
商品推荐第二子模块,配置成当待推荐用户在热点区域中时,基 于所述多个第一样本用户的历史购物行为和热点区域中商品的销量进行商品推荐;
其中所述多个第一样本用户包括待推荐用户。
在根据本公开一些实施例的商品推荐设备中,热点区域获取模块包括:
位置获取模块,配置成获取多个第二样本用户的位置坐标;
聚类模块,配置成采用聚类算法,基于所述多个第二样本用户的位置坐标对所述多个第二样本用户进行聚类,得到至少一个第二样本用户类,
热点区域确定模块,配置成根据所述至少一个第二样本用户类确定热点区域,其中与每一个第二样本用户类对应的热点区域的中心点坐标
Figure PCTCN2019079433-appb-000001
和半径r通过下式确定:
Figure PCTCN2019079433-appb-000002
Figure PCTCN2019079433-appb-000003
其中,n为与该热点区域对应所述第二样本用户类中用户的个数,且(x i,y i)为与该热点区域对应所述第二样本用户类中第i个用户的位置坐标,其中i=1,...,n。
在根据本公开一些实施例的商品推荐设备中,商品推荐第一子模块包括:
数据获取模块,配置成获取所述多个第一样本用户的历史购物行为数据;
评分矩阵确定模块,配置成根据所述历史购物行为数据,确定第一样本用户-商品的评分矩阵;
预测评分矩阵确定模块,配置成根据所述第一样本用户-商品的评分矩阵,利用基于矩阵分解的协同过滤算法确定所述第一样本用户-商品的预测评分矩阵;
商品推荐第三子模块,配置成根据所述第一样本用户-商品的预测评分矩阵,向所述待推荐用户进行商品推荐。
根据本公开的第二方面,提供一种商品推荐系统,包括商品推荐设备以及与商品推荐设备通信连接的用户终端,其中:
用户终端被配置用于向商品推荐设备发送待推荐用户的位置信息;
商品推荐设备被配置用于接收待推荐用户的位置信息,根据待推荐用户的位置信息进行商品推荐,以及将所推荐的商品信息发送给待推荐用户;并且
用户终端进一步被配置用于从商品推荐设备接收并展示所推荐的商品信息。
根据本公开的第三方面,提供一种商品推荐方法,包括:
接收待推荐用户的位置信息;
根据待推荐用户的位置信息,进行商品推荐;
将所推荐的商品信息发送给待推荐用户。
在根据本公开一些实施例的商品推荐方法中,所述根据待推荐用户的位置信息进行商品推荐的步骤包括:
获取热点区域;
根据待推荐用户的位置信息,确定所述待推用户是否在热点区域中;
当待推荐用户不在热点区域中时,基于多个第一样本用户的历史购物行为进行商品推荐,其中所述多个第一样本用户包括待推荐用户;
当待推荐用户在热点区域中时,基于所述多个第一样本用户的历史购物行为和热点区域中商品的销量进行商品推荐。
在根据本公开一些实施例的商品推荐方法中,所述获取热点区域的步骤包括:
获取多个第二样本用户的位置坐标;
采用聚类算法,基于所述多个第二样本用户的位置坐标对所述多个第二样本用户进行聚类,得到至少一个第二样本用户类,
根据所述至少一个第二样本用户类确定热点区域,其中与每一个第二样本用类对应的热点区域的中心点
Figure PCTCN2019079433-appb-000004
和半径r通过下式确定:
Figure PCTCN2019079433-appb-000005
Figure PCTCN2019079433-appb-000006
其中,n为与该热点区域对应所述第二样本用户类中用户的个数, 且(x i,y i)为与该热点区域对应所述第二样本用户类中第i个用户的位置坐标,其中i=1,...,n。
在根据本公开一些实施例的商品推荐方法中,所述聚类算法包括OPTICS算法或DBSCAN算法。
在根据本公开一些实施例的商品推荐方法中,所述基于多个第一样本用户的历史购物行为进行商品推荐的步骤包括:
获取所述多个第一样本用户的历史购物行为数据;
根据所述历史购物行为数据,确定第一样本用户-商品的评分矩阵;
根据所述第一样本用户-商品的评分矩阵,利用基于矩阵分解的协同过滤算法确定所述第一样本用户-商品的预测评分矩阵;
根据所述第一样本用户-商品的预测评分矩阵预测,向所述待推荐用户进行商品推荐。
在根据本公开一些实施例的商品推荐方法中,所述多个第一样本用户为m个第一样本用户U1,...,Um,所述多个第一样本用户的历史购物行为涉及n种商品V1,...,Vn,并且所述根据所述历史购物行为数据,确定第一样本用户-商品的评分矩阵的步骤包括:
根据所述历史购物行为数据,确定每个第一样本用户Ui对商品V1,...,Vn的评分,其中在第一样本用户Ui的历史购物行为不涉及商品Vj的情况下,Ui对Vj的评分为空值,其中i=1,...,m,j=1,...,n;
根据所述每个第一样本用户Ui对商品V1,...,Vn的评分,确定第一样本用户-商品的评分矩阵A mxn,其中矩阵A的元素A(i,j)表示用户Ui对商品Vj的评分。
在根据本公开一些实施例的商品推荐方法中,所述根据所述第一样本用户-商品的评分矩阵,利用基于矩阵分解的协同过滤算法确定所述第一样本用户-商品的预测评分矩阵的步骤包括:
基于矩阵分解的协同过滤算法,获得矩阵U m×k和V n×k,使得A≈U×V T,k<<m,n,
通过下式确定第一样本用户-商品的预测评分矩阵A′
A′=U×V T
其中矩阵A′的元素A′(i,j)表示用户Ui对商品Vj的预测评分。
在根据本公开一些实施例的商品推荐方法中,所述根据所述第一样本用户-商品的预测评分矩阵,向所述待推荐用户进行商品推荐的步 骤包括:
根据在第一样本用户-商品的预测评分矩阵A′中的待推荐用户对商品U1,...,Um的评分,向所述待推荐用户进行商品推荐。
在根据本公开一些实施例的商品推荐方法中,所述根据待推荐用户的位置信息确定所述待推荐用户是否在热点区域中的步骤包括:
计算待推荐用户的位置与热点区域的中心点之间的欧式距离;
将所计算的距离与热点区域的半径进行比较:若所述距离大于半径,则用户不在热点区域中;否则,用户在热点区域中。
在根据本公开一些实施例的商品推荐方法中,所述当用户在热点区域中时基于所述多个第一样本用户的历史购物行为和热点区域中商品的销量进行商品推荐的步骤包括:
对所述热点区域中的商品按照销量从高到低进行排序,得到排序在前的N种商品,其中N为大于或者等于1的整数,
获取所述多个第一样本用户的历史购物行为数据;
根据所述历史购物行为数据,确定第一样本用户-商品的评分矩阵;
针对所述第一样本用户-商品的评分矩阵,利用基于矩阵分解的协同过滤算法确定第一样本用户-商品的预测评分矩阵;
根据所述第一样本用户-商品的预测评分矩阵以及所述热点区域中排序在前的N种商品,进行商品推荐。
在根据本公开一些实施例的商品推荐方法中,所述根据所述第一样本用户-商品的预测评分矩阵以及所述热点区域中排序在前的N种商品进行商品推荐的步骤包括:
根据所述第一样本用户-商品的预测评分矩阵,获得至少一种第一推荐商品,且根据热点区域中排序在前的N种商品获得至少一种第二推荐商品;以及
将所述第一和第二推荐商品以固定的或随机的优先顺序推荐给待推荐用户。
在根据本公开一些实施例的商品推荐方法中,所述基于矩阵分解的协同过滤算法包括基于交叉最小二乘法矩阵分解的协同过滤算法或基于梯度下降法矩阵分解的协同过滤算法。
在根据本公开一些实施例的商品推荐方法中,在根据待推荐用户的位置信息进行商品推荐之后且在所述将所推荐的商品信息发送给待 推荐用户之前,还包括:
获取所述待推荐用户已放入购物车的商品信息;
在所推荐的商品信息中包含已放入购物车的商品的情况下,将已放入购物车的商品从所推荐的商品信息中删除。
在根据本公开一些实施例的商品推荐方法中,所述获取所述待推荐用户已放入购物车的商品信息的步骤包括:利用购物车上的商品识别装置获取已放入购物车的商品的商品信息。
根据本公开的第四方面,提供一种计算设备,包括:处理器;以及存储器,其上存储有计算机可读指令,当所述计算机可读指令被处理器执行时,使所述计算设备执行根据本公开一些实施例的商品推荐方法。
根据本公开的第五方面,提供一种计算机可读存储介质,包括存储在其上的计算机可读指令,所述计算机可读指令在被执行时实现根据本公开一些实施例的商品推荐方法。
附图说明
图1示出根据本公开一些实施例的商品推荐方法的流程图;
图2示出根据本公开一些实施例的商品推荐方法中根据位置信息进行商品推荐的流程图;
图3示出根据本公开一些实施例的商品推荐方法中获取热点区域的流程图;
图4形成根据本公开一些实施例的商品推荐方法中基于多个第一样本用户的历史购物行为进行商品推荐的流程图;
图5示出根据本公开一些实施例的商品推荐方法中基于多个第一样本用户的历史购物行为和热点区域中商品的销量进行商品推荐的流程图;
图6示出根据本公开另一些实施例的商品推荐方法的流程图;
图7示意性示出根据本公开一些实施例的商品推荐设备的结构框图;以及
图8示意性示出根据本公开一些实施例的商品推荐系统的结构图。
具体实施方式
为使本公开要解决的技术问题、技术方案和优点更加清楚,下面将结合附图及具体实施例进行详细描述。
图1示出本公开一些实施例的商品推荐方法的流程图。在线下商店或超市中,可以设置一种商品推荐设备以用于执行图1所示的商品推荐方法,该商品推荐设备可以是具备数据收发、处理能力的任意计算设备,例如服务器、计算机等。如图1所示,根据本公开一些实施例的商品推荐方法包括以下步骤S101-S103。
S101,接收待推荐用户的位置信息。
上述位置信息可以是用户在商场或超市购物时所处的位置坐标,该位置坐标例如可采用平面(二维)或空间(三维)直角坐标系中的坐标点来表示,例如在商场仅包含一层时,可以采用平面坐标系,而如果商场较大,包含多层,这时需要空间坐标系来唯一表示商场中的各个位置。当然,在本公开中,位置坐标也可以采用其他表示方式,例如极坐标等等。下文中,将以仅包含一层的商场为例进行说明。
用户可以通过用户终端(例如手机或安装在购物车上的电子设备)中集成的定位功能(例如GPS模块)来对自身进行定位,并通过该终端将位置信息发送至商场或超市中的商品推荐设备。由于用户终端与用户的距离非常近,因此,用户终端的位置信息可视为是用户的位置信息。用户终端可每隔预设时间向服务器发送位置信息。
在一些实施例中,在步骤S101中,还可以包括接收用户的身份信息或登录信息。身份信息或登录信息包括用于唯一标识用户的用户ID。当然,登录信息还可包括用户照片、姓名、生日、年龄、性别、职业、购物累计次数等多个用户基本信息中的至少之一。商品推荐设备接收用户登录信息的方式可以包括:首先,用户终端通过下述方式获取登录信息:接收用户手动输入用户ID、扫描二维码或者扫描芯片卡、利用自带的摄像头对用户进行面部识别等等;随后,用户终端将获取到的登录信息通过有线或无线网络发送给商品推荐设备。
在一些实施例中,商场中的商品推荐设备或其他服务器中存储有与用户身份信息或用户ID对应的用户记录或会员信息。一般地,用户首次在商场进行购物时,可以进行登记成为会员并录入会员信息;登记之后,用户会获得一张会员卡,由于每张会员卡均对应唯一的卡号,所以可将会员卡的卡号作为用户ID;随后,将用户ID录入服务器中, 或者直接将用户的登记信息录入服务器中且由服务器生成用户ID,返回给用户,以供用户在商场购物时使用。用户ID例如可以通过一维码、二维码或卡号等方式表示。会员数据库中的会员信息可以是在用户初始注册时录入的信息,也可以是通过对用户相关数据(例如销售数据)进行记录和分析后得出的信息。会员信息可以包括用户的个人信息,例如生日、年龄、性别、民族、职业等。会员信息还可以包括用户历史购物行为,例如所购商品、购买次数、购买偏好等。
S102,根据待推荐用户的位置信息,进行商品推荐。
在接收到用户位置信息之后,商品推荐设备可以根据用户所在位置,向其推荐该位置附近的商品,例如促销商品或畅销商品等。此外,也可以根据用户的历史购物记录推荐该位置附近的商品。
在一些实施例中,可以根据用户所在的不同位置,采用不同的不同方式进行商品推荐。例如,可以将线性购物场所(例如商场或超市)换分成不同区域,然后根据用户所在的不同区域的商品特性,向用户推荐用户所需商品。例如,如下文详细所述,可以将商场划分成热点区域(即热销区域或人流密集区域)和非热点区域,然后依据用户是否在热点区域,而采用不同的算法进行商品推荐,即在非热点区域可以采用协同过滤算法,而在热点区域则采用协同过滤算法结合商品销量topN算法。
这种基于位置的推荐方法从而更好地贴近用户的需求,进一步提高商品推荐的精确性和购物的便利性。
S103,将所推荐的商品信息发送给待推荐用户。
在完成商品推荐之后,可以将推荐的商品信息通过有线或无线网络(例如wifi等)发送至用户终端,从而告知用户所推荐的商品的相关信息,例如商品名称、在商场中的具体位置等,以供用户参考。
图2示出根据本公开一些实施例的商品推荐方法中根据位置信息进行商品推荐的流程图。如图2所示,图1所示的步骤S102-根据用户的位置信息进行商品推荐的步骤可以包括下述步骤S1021-1024。
S1021,确定热点区域。
在一些实施例中,可以预先将商场或超市的购物区域划分或分类成若干不同的区域,然后根据用户所处的不同区域采用不同的商品推荐方法。例如,步骤S1021中所述的热点区域表示商场中人流密集的 区域或购物者停留时间较长的区域,可以为预先确定的区域,例如,热点区域可直接根据商场中商品的布局情况进行预先设置;或者也可通过在一段时间内用户的位置信息或运动轨迹数据设定,例如可以通过将一定时间内位置变化较小的用户根据其位置坐标进行聚类来确定热点区域,即这类用户所在的区域即为热点区域,详见下文结合图3的描述。
S1022,根据待推荐用户的位置信息,确定所述待推荐用户是否在热点区域中。
S1023,当待推荐用户不在热点区域中时,基于多个第一样本用户的历史购物行为进行商品推荐,其中所述多个第一样本用户包括待推荐用户。
步骤S1023的目的是根据目标用户或待推荐用户和商场中其他用户购物行为来预测当前用户的购物行为,从而向目标用户推荐商品。该推荐方式基于下述思想,不同用户的商品购买情况具有一定的相似性,所以结合所有用户的购买习惯进行推荐具有一定的参考性。
第一样本用户可以包括在商场中登记注册为会员的所有用户。可选地,第一样本用户也可以仅包括部分会员用户,排除那些没有购物记录或购物记录较少的会员用户,以简化计算量并提高推荐的准确性。第一样本用户可定时更新,以便与时俱进、更加符合用户购物行为的实际情况。
用户购物行为是指用户在商场或超市中购物时与商品购买相关的行为或动作,例如浏览商品、查看商品、将商品放入购物车等等。一般地,可以将用户购物行为大致划分为四种:未查看商品、仅查看商品、查看商品并放入购物车但最终未购买、以及购买。以上划分方式仅用于示例,在实际应用中也可采用其他方式来标识用户的购物行为。
在一些实施例中,可以将上述购物行为标识转化为数字,这些数字可以看作用户对相应商品的评分,于是可以得到多个第一样本用户U1-Um对多个商品V1-Vn的评分。对于每个样本用户(其中包含待推荐用户或目标用户),可以根据上述评分,通过基于矩阵分解的协同过滤算法来预测出该用户对未评分(即用户购物行为是“空值”的)商品的评分,进而将预测评分高的商品推荐给该用户。关于基于矩阵分解的协同过滤算法进行商品预测的具体方式,请参见下文结合图4 的描述。
S1024,当用户位置在的热点区域中时,基于所述多个第一样本用户的历史购物行为和热点区域中商品的销量进行商品推荐。
在一个实施例中,当待推荐用户在所述热点区域内时,可以基于热点区域中商品的销量并结合所述步骤S1023中的多个第一样本用户的历史购物行为向待推荐用户推荐商品。一方面,在现实世界里,由于多数用户具有相似喜好,购买的商品具有相似性,而销量较大的商品可在很大程度上表明多数用户都需要,而且这样的商品可能具有区别于其他商品的优点,比如物美价廉、新品促销、独家销售等等,因此按照销量进行商品推荐是可行的;而且当用户位于热点区域内时,向用户推荐同样位于热点区域内的商品,便于用户方便拿取商品,可有效避免用户绕行太远去拿取商品,进一步方便用户的购物需求。另一方面,由于相对于整个商场而言,热点区域仅仅是其局部区域,因而处于热点区域的商品是有限的,因此单纯依据热点区域内的商品销量进行推荐可能是片面的,无法充分满足用户的需求。于是,当待推荐用户位于热点区域中时,可以在考虑区域内热销商品的基础上结合步骤S1023中的基于多个第一样本用户(其中包括待推荐用户)的历史购物行为(即所有或部分用户对商场内所有商品的购物行为或评分)的推荐方式进行商品推荐,以更精确地向用户推荐其所需要、喜爱的商品。
图3示出根据本公开一些实施例的商品推荐方法中获取热点区域的流程图。如上文所述,热点区域表示商场中人流密集的区域或购物者停留时间较长的区域。因此,除了上文所述的根据商品布局或历史购物信息预先确定热点区域之外,还可以根据商场中当前正在购物的所有顾客或消费者所在的位置来确定热点区域,例如顾客密集的区域,即位置点集中的区域,可以定义为热点区域。于是,可以将位置点集中在某一区域中的顾客归为一类,则这类顾客所在的区域就可以确定为热点区域。因此,可以将当前商场中的所有用户的位置坐标作为样本点,通过聚类算法,将这些样本点(或对应的用户)进行聚类,得到用户(位置点)的一个或多个类(即位置点相互靠近的用户的集合),所述一个或多个类的用户所在的区域就是热点区域。
如图3所示,图2所示的获取热点区域的步骤S1021包括下述步 骤S1021a-S1021c。
S1021a,获取多个第二样本用户的位置坐标。
第二样本用户可以指当前在商场的购物区域中所有顾客,其中购物区域为商场中除去商场入口、出口和电梯区域的区域。可选地,第二样本用户也可以是正在商场中购物的部分顾客,例如其可以限定为在一定时间段内位置变化较小的用户(即在某个区域中停留时间较长的顾客),这样其所停留的区域更接近热点区域。例如可以根据顾客的运动轨迹来确定第二样本用户。
与图1所示的步骤S101类似,第二样本用户也可以通过用户终端(例如手机或安装在购物车上的电子设备)中集成的定位功能来对自身进行定位,并通过该终端将位置信息发送至商场或超市中的商品推荐设备。上述位置信息可以是用户在商场或超市购物时所处的位置坐标,该位置坐标例如可采用平面直角坐标系中的坐标(x,y)来表示。
S1021b,采用聚类算法,以所述多个第二样本用户的位置坐标为样本点,对所述多个第二样本用户进行聚类,得到第二样本用户的一个或多个类别,
在一些实施例中,可以采用聚类算法OPTICS(Ordering Points to identify the clustering structure)来对第二样本用户进行聚类,以获得一个或多个第二样本用户的类。OPTICS聚类算法是基于密度的聚类算法,目标是将空间中的数据按照密度分布进行聚类。这样经过基于位置坐标点的密度的聚类所得到的每个用户类所在的区域就可视为一个热点区域,这样获得的热点区域能够有效反映用户在该热点区域内出现的频率,用户出现的频率越高,说明用户在该热点区域购物的可能性越大,将在该热点区域的商品推荐给用户,可提高商品推荐的准确性。可选地,上述步骤S1021b也可以采用其他聚类算法,例如DBSCAN(Density-Based Spatial Clustering of Applications with Noise)。在一些实施例中,采用聚类算法对第二样本用户进行聚类的具体过程如下:
输入:数据样本D,其中数据样本D为第二样本用户的坐标点;
初始化:所有坐标点的可达距离和核心距离为MAX,领域半径ε和给定点在ε邻域内成为核心对象的最少领域点数MinPts,其中MAX、ε和MinPts可根据实际情况设置;
步骤1,建立两个队列,分别为有序队列(待处理数据)和结果队 列(已处理数据),其中有序队列用来存储核心对象及其该核心对象的直接可达对象,并按可达距离升序排列;结果队列用来存储样本点的输出次序;
步骤2,如果样本D中所有样本点数据全部处理完,则算法结束;否则从D中选择一个未处理且为核心对象的点,将该核心对象的点放入结果队列,且将该核心点的直接密度可达点放入有序队列并按可达距离升序排列;
步骤3,如果有序序列为空,则回到步骤2,否则从有序队列中取出第一个点;
3.1判断该点是否为核心点,不是则回到步骤3,是的话则将该点存入结果队列,如果该点不在结果队列;
3.2该点是核心点的话,找到其所有直接密度可达点,并将这些点放入有序队列,且将有序队列中的点按照可达距离重新排序,如果该直接密度可达点已经在有序队列中且新的可达距离小于旧的可达距离,则用新的可达距离取代旧的可达距离;
3.3重复步骤3,直至有序队列为空;
步骤4,从结果队列中按顺序取出点,如果该点的可达距离不大于给定半径ε,则该点属于当前类别,否则至步骤5;
步骤5,如果该点的核心距离大于给定半径ε,则该点为噪声,可以忽略,否则该点属于新的类,跳至步骤1;
步骤6,结果队列遍历结束,则算法结束。
S1021c,根据所述一个或多个第二样本用户类确定热点区域,其中与每一个第二样本用类对应的热点区域的中心点
Figure PCTCN2019079433-appb-000007
和半径r通过下式确定:
Figure PCTCN2019079433-appb-000008
Figure PCTCN2019079433-appb-000009
其中:n为与该热点区域对应的第二样本用户类中用户的个数,且(x i,y i)为与该热点区域对应所述第二样本用户类中第i个用户的位置坐标,其中i=1,...,n。
如上述步骤S1021c中的公式(1)和(2)所示,所得到的热点区 域可以为圆形区域。于是,在判断待推荐用户是否在热点区域中时,可以通过计算待推荐用户的位置与所述热点区域的中心点
Figure PCTCN2019079433-appb-000010
之间的欧式距离d并将其与所述热点区域的半径r进行比较来实现:若d>r,则用户不在热点区域中;若d<r,则用户在热点区域中。可选地,除了上述圆形热点区域之外,也可以通过其他方式来限定不同形状的热点区域。
图4示出根据本公开一些实施例的商品推荐方法中基于第一样本用户的历史购物行为进行商品推荐的流程图。如图4所示,图2所示的步骤S1023进一步包括下述步骤S1023a-S1023d。
S1023a,获取多个第一样本用户的历史购物行为数据。
如上文所述,在一些实施例中,第一样本用户的历史购物行为数据可采用如下方式设置:
当第一样本用户未查看商品时,第一样本用户对该商品的购物行为数据标识为空值;
当第一样本用户只查看商品时,第一样本用户对该商品的购物行为数据标识为“查看”;
当第一样本用户查看商品并将该商品放入购物车但最终未购买时,第一样本用户对该商品的购物行为数据标识为“放入购物车”;
当第一样本用户购买商品时,第一样本用户对该商品的购物行为数据标识为“购买”。
S1023b,根据所述历史购物行为数据,确定第一样本用户-商品评分矩阵。
为处理方便的目的,可以将上述历史购物行为数据转化为数字表示,这些数字可以看作用户对相应商品的评分。在一些实施例中,可以根据表1确定第一样本用户的历史购物行为与其对商品的评分之间的对应关系。可选地,也可以通过其他方式定义购物行为与商品评分之间的关系。
表1:购物行为与商品评分的对应关系表
用户购物行为数据 用户对商品的评分
未查看(且未购买) 空值(缺失)
查看(但未放入购物车且未购买) 1
放入购物车(但未购买) 2
购买 3
这样,可以得到如表2所示的多个第一样本用户U1-U5对多个商品V1-V4的评分表。表2为一个示例性的第一样本用户-商品评分表,其中,U1-U5分别表示不同的第二样本用户,V1-V4分别表示不同的商品,表中的数值为第二样本用户对相应商品的评分,其中“-”表示空值。
表2:第一样本用户-商品评分表
Figure PCTCN2019079433-appb-000011
为了计算方便起见,上述评分表可以转换成评分矩阵的形式。因此,对于m个第一样本用户U1,...,Um和n种商品V1,...,Vn而言,可以根据第一样本用户的历史购物行为数据,通过上述表1所示的评分策略,得到一个m×n的第一样本用户-商品评分矩阵A。矩阵A中的元素A(i,j)表示用户Ui对商品Vj的评分,该评分可以代表用户Ui对商品Vj的感兴趣程度,分数越高表示该用户U i对该商品Vj越感兴趣。如果评分为“空值”或缺失(用“-”表示),则表示用户Ui针对商品Vj的购物行为是“未查看”,即第一样本用户Ui的历史购物行为记录中没有关于Vj的历史购物行为(因此无法评分)。因此,矩阵A中这些缺失值就是我们想要确定的用户Ui针对商品Vj的预测评分,预测评分的过程即为矩阵A的空值补全过程。
S1023c,根据所述第一样本用户-商品的评分矩阵,利用基于矩阵分解的协同过滤算法确定所述第一样本用户-商品的预测评分矩阵。
在本公开中,通过将第一样本用户-商品的评分矩阵A m×n的空值或 缺失补全即确定所有缺失处的预测评分而得到的无缺失值的矩阵可以称为第一样本用户-商品的预测评分矩阵A′。为了得到第一样本用户-商品的预测评分矩阵A′,可以采用基于交叉最小二乘法(ALS,alternative least squares)矩阵分解的协同过滤算法。
一般地,在基于ALS矩阵分解的协同过滤算法中,由于包含缺失值的评分矩阵A通常是低秩的,因此m×n的评分矩阵A可以用两个小矩阵U m×k和V n×k的乘积来近似:A≈U×V T,k<<m,n。按照基于ALS矩阵分解的协同过滤算法,可以求出矩阵U和V,从而可以利用U和V的乘积来对用户-商品的评分矩阵进行还原,即对矩阵中原本存在的缺失值进行预测,得到了第一样本用户-商品的预测评分矩阵A′=U×V T。关于基于ALS矩阵分解的协同过滤算法,可以利用计算引擎Spark进行模型训练和调优。
可选地,也可以采用基于梯度下降法的矩阵分解协同过滤算法来确定所述第一样本用户-商品的预测评分矩阵。
S1023d,根据所述第一样本用户-商品的预测评分矩阵,向所述待推荐用户进行商品推荐。
在确定了第一样本用户-商品的预测评分矩阵之后,可以确定所有第一样本用户针对商场内所有商品的评分,而由于第一样本用户包含待推荐用户,因此可以从预测评分矩阵中得知待推荐用户对商场内所有商品的预测评分。因此,可以根据预测评分的高低来向用户推荐商品。例如,将用户对各商品预测评分按照从高到低的顺序排列,选择排序在前M种商品,M为正整数。
在本实施例中,由于预测评分反映了该用户对相应商品的感兴趣程度,因此根据用户对各种商品的预测评分来确定推荐商品,使得推荐方案更精确地契合用户的需求和愿望,增强了个性化体验。
图5示出根据本公开一些实施例的商品推荐方法中的基于多个第一样本用户的历史购物行为和热点区域中商品的销量进行商品推荐的流程图。如图5所示,图3中的步骤S1024包括下述步骤:
S1024a,对所述热点区域中的商品按照销量从高到低进行排序,得到排序在前的N种商品,其中N为大于或者等于1的整数;
S1024b,获取第一样本用户的历史购物行为数据;
S1024c,根据所述历史购物行为数据,确定第一样本用户-商品的 评分矩阵;
S1024d,基于所述第一样本用户-商品的评分矩阵,利用基于矩阵分解的协同过滤算法确定第一样本用户-商品的预测评分矩阵;
S1024e,根据所述第一样本用户-商品的预测评分矩阵以及热点区域中排序在前的N种商品,进行商品推荐。
在本实施例中,当用户在所述热点区域内时,采用两种方式组合确定推荐商品。第一种方式为:根据位于热点区域的商品的销量(即TopN算法)来进行商品推荐;第二种方式为:根据第一样本用户(例如,商场中所有有购物行为记录的用户或会员)的历史购物行为进行商品推荐,这与用户不在热点区域中时的方式相同,参见图4所示的流程图。上述这种组合方式既考虑了用户的特征(本人和其他用户的历史购物习惯),又考虑了商品的特征(商品的销量和位置),从而更好地贴近用户的需求,进一步提高商品推荐的精确性和购物的便利性。
关于两种不同的推荐方式如何组合地运用,可以采取如下措施:例如可以分别利用协同过滤算法和topN算法获得若干推荐商品,然后可以将所得到不同的推荐商品以不同的优先次序推荐给用户,将这些推荐商品固定或随机地置于不同的推荐位(即设置推荐优先次序)以供用户自主进行选择。关于如何设置不同算法得到不同推荐商品的推荐位的问题,可以先随机或固定设置,之后根据用户选择(点击)的具体情况进行动态调整。例如,如果topN算法所得的推荐商品点击率较高,则将此类商品置于靠前的推荐位,进行优先推荐;反之将协同过滤算法获得的推荐商品进行优先推荐。
具体地,在一些实施例中,图5所示的步骤S1024e-根据所述第一样本用户-商品的预测评分矩阵以及热点区域中排序在前的N种商品进行商品推荐可以包括:
首先,根据所述第一样本用户-商品的预测评分矩阵,获得至少一种第一推荐商品,且根据热点区域中排序在前的N种商品获得至少一种第二推荐商品;
其次,将所述第一和第二推荐商品以固定的或随机的优先顺序推荐给待推荐用户。
图6示出根据本公开另一些实施例的商品推荐方法的流程图。与 图1相比,图6中所述的商品推荐方法,在步骤S102之后且在步骤S103之前,进一步包括下述步骤S104-S105。
S104,获取所述待推荐用户已放入购物车的商品信息。
举例来说,购物车上设置有摄像头或者扫描设备。当用户将商品放入购物车时,购物车上的摄像头或者扫描设备获取到商品的商品信息,该商品信息可以包括商品ID,还可以包括商品名称、商品所属类别、商品售价等。购物车将获取到的商品信息会进一步发送给商品推荐设备。
S105,在所推荐的商品信息中包含已放入购物车的商品的情况下,将已放入购物车的商品从所述推荐商品信息中删除。
当服务器确定推荐商品之后,可以根据获得的商品信息在推荐商品中进行查询,若推荐商品信息中包括了商品信息所对应的商品(以下称为B商品),说明此时用户已经将推荐商品中的B商品放入购物车,为了避免用户参考推荐商品重复购买A商品,可将推荐商品信息中包括的B商品删除,以获得新的推荐商品信息。
图7示意性示出根据本公开一些实施例的商品推荐设备的结构框图。根据本发明一些实施例的商品推荐方法可以通过如图7所示的商品推荐设备完成。如图7所示,该商品推荐设备700包括:
信息获取模块710,配置成接收待推荐用户的位置信息;
商品推荐模块720,配置成根据待推荐用户的位置信息,进行商品推荐;
信息发送模块730,配置成将所推荐的商品信息发送给待推荐用户。
在根据本公开一些实施例的商品推荐设备中,商品推荐模块包括:
热点区域获取模块,配置成获取热点区域;
位置判断模块,配置成根据待推荐用户的位置信息,确定所述待推用户是否在热点区域中;
商品推荐第一子模块,配置成当待推荐用户不在热点区域中时,基于多个第一样本用户的历史购物行为进行商品推荐;
商品推荐第二子模块,配置成当待推荐用户在热点区域中时,基于所述多个第一样本用户的历史购物行为和热点区域中商品的销量进行商品推荐;
其中所述多个第一样本用户包括待推荐用户。
在根据本公开一些实施例的商品推荐设备中,热点区域获取模块包括:
位置获取模块,配置成获取多个第二样本用户的位置坐标;
聚类模块,配置成采用聚类算法,基于所述多个第二样本用户的位置坐标对所述多个第二样本用户进行聚类,得到至少一个第二样本用户类,
热点区域确定模块,配置成根据所述至少一个第二样本用户类确定热点区域,其中与每一个第二样本用户类对应的热点区域的中心点坐标
Figure PCTCN2019079433-appb-000012
和半径r通过下式确定:
Figure PCTCN2019079433-appb-000013
Figure PCTCN2019079433-appb-000014
其中,n为与该热点区域对应所述第二样本用户类中用户的个数,且(x i,y i)为与该热点区域对应所述第二样本用户类中第i个用户的位置坐标,其中i=1,...,n。
在根据本公开一些实施例的商品推荐设备中,商品推荐第一子模块包括:
数据获取模块,配置成获取所述多个第一样本用户的历史购物行为数据;
评分矩阵确定模块,配置成根据所述历史购物行为数据,确定第一样本用户-商品的评分矩阵;
预测评分矩阵确定模块,配置成根据所述第一样本用户-商品的评分矩阵,利用基于矩阵分解的协同过滤算法确定所述第一样本用户-商品的预测评分矩阵;
商品推荐第三子模块,配置成根据所述第一样本用户-商品的预测评分矩阵,向所述待推荐用户进行商品推荐。
根据本公开的第二方面,提供一种商品推荐系统,包括商品推荐设备以及与商品推荐设备通信连接的用户终端,其中:
用户终端被配置用于向商品推荐设备发送待推荐用户的位置信息;
商品推荐设备被配置用于接收待推荐用户的位置信息,根据待推荐用户的位置信息进行商品推荐,以及将所推荐的商品信息发送给待推荐用户;并且
用户终端进一步被配置用于从商品推荐设备接收并展示所推荐的商品信息。
图8示意性示出了根据本公开一些实施例的商品推荐系统的结构图。如图8所示,该商品推荐系统包括商品推荐设备810以及与商品推荐设备810通信连接的用户终端820,其中:
用户终端820被配置用于向商品推荐设备发送待推荐用户的位置信息;
商品推荐设备810被配置用于接收待推荐用户的位置信息,根据待推荐用户的位置信息进行商品推荐,以及将所推荐的商品信息发送给待推荐用户;并且
用户终端820进一步被配置用于从商品推荐设备810接收并展示所推荐的商品信息。
本发明还涉及一种或多种计算机存储介质,其上存储有计算机程序,当计算机程序在被执行时实现如上所述的根据本发明一些实施例的商品推荐方法。此外,本发明还涉及一种计算设备,包括:处理器;以及存储器,其上存储有计算机可读指令,当所述计算机可读指令被处理器执行时,使所述计算设备执行如上所述的根据本发明一些实施例的商品推荐方法。
在本说明书的描述中,术语“一个实施例”、“一些实施例”、“示例”、“具体示例”、或“一些示例”等的描述意指结合该实施例或示例描述的具体特征、结构、材料或者特点被包含于本发明的至少一个实施例或示例中。在本说明书中,对上述术语的示意性表述不必针对的是相同的实施例或示例。而且,描述的具体特征、结构、材料或者特点可以在任一个或多个实施例或示例中以合适的方式结合。此外,在不相互矛盾的情况下,本领域的技术人员可以将本说明书中描述的不同实施例或示例以及不同实施例或示例的特征进行结合和组合。
流程图中或在此以其他方式描述的任何过程或方法描述可以被理解为,表示包括一个或多个用于实现定制逻辑功能或过程的步骤的可执行指令的代码的模块、片段或部分,并且本发明的优选实施方式的 范围包括另外的实现,其中可以不按所示出或讨论的顺序(包括根据所涉及的功能按基本同时的方式或按相反的顺序)来执行功能,这应被本发明的实施例所属技术领域的技术人员所理解。
在流程图中表示或在此以其他方式描述的逻辑和/或步骤,例如,可以被认为是用于实现逻辑功能的可执行指令的定序列表,可以具体实现在任何计算机可读介质中,以供指令执行系统、装置或设备(如基于计算机的系统、包括处理器的系统或其他可以从指令执行系统、装置或设备取指令并执行指令的系统)使用,或结合这些指令执行系统、装置或设备而使用。就本说明书而言,“计算机可读介质”可以是任何可以包含、存储、通信、传播或传输程序以供指令执行系统、装置或设备或结合这些指令执行系统、装置或设备而使用的装置。计算机可读介质的更具体的示例可以例如包括以下各项:具有一个或多个布线的电连接部(电子装置)、便携式计算机盘盒(磁装置)、随机存取存储器(Random Access Memory)、只读存储器(Read Only Memory),可擦除可编辑只读存储器(Erasable Programmable Read Only Memory)或闪速存储器、光纤装置、以及便携式光盘只读存储器(Compact Disc Read Only Memory)。另外,计算机可读介质甚至可以是可在其上打印所述程序的纸或其他合适的介质,因为可以例如通过对纸或其他介质进行光学扫描,接着进行编辑、解译或必要时以其他合适方式进行处理来以电子方式获得所述程序,然后将其存储在计算机存储器中。
应当理解,本发明的各部分可以用硬件、软件、固件或它们的组合来实现。在上述实施方式中,多个步骤或方法可以用存储在存储器中且由合适的指令执行系统执行的软件或固件来实现。例如,如果用硬件来实现,则可用本领域公知的下列技术中的任一项或它们的组合来实现:具有用于对数据信号实现逻辑功能的逻辑门电路的离散逻辑电路、具有合适的组合逻辑门电路的专用集成电路、可编程门阵列(Programmable Gate Array)、现场可编程门阵列(Field Programmable Gate Array)等。
本技术领域的普通技术人员可以理解上述实施例方法的全部或部分步骤可以通过程序指令相关的硬件完成,所述程序可以存储于一种计算机可读存储介质中,该程序在执行时,包括执行方法实施例的步骤之一或其组合。
此外,在本发明各个实施例中的各功能单元可以集成在一个处理模块中,也可以是各个单元单独物理存在,也可以两个或两个以上单元集成在一个模块中。上述集成的模块既可以采用硬件的形式实现,也可以采用软件功能模块的形式实现。所述集成的模块如果以软件功能模块的形式实现并作为独立的产品销售或使用时,也可以存储在一个计算机可读取存储介质中。
应当注意,在权利要求书中,动词“包括/包含”及其变体的使用并没有排除存在权利要求中未陈述的元件或步骤。措词“一”或“一个”并没有排除多个。
尽管已经示出和描述了本发明的特定实施例,但是对于本领域技术人员显然的是,可以在不脱离发明的情况下在其更宽的方面做出若干改变和修改,因此,所附权利要求书应当在其范围内包含所有这样的改变和修改,如同落入本发明的真实精神和范围之内。

Claims (20)

  1. 一种商品推荐设备,包括:
    信息获取模块,配置成接收待推荐用户的位置信息;
    商品推荐模块,配置成根据待推荐用户的位置信息,进行商品推荐;
    信息发送模块,配置成将所推荐的商品信息发送给待推荐用户。
  2. 如权利要求1所述的商品推荐设备,其中所述商品推荐模块包括:
    热点区域获取模块,配置成获取热点区域;
    位置判断模块,配置成根据待推荐用户的位置信息,确定所述待推用户是否在热点区域中;
    商品推荐第一子模块,配置成当待推荐用户不在热点区域中时,基于多个第一样本用户的历史购物行为进行商品推荐;
    商品推荐第二子模块,配置成当待推荐用户在热点区域中时,基于所述多个第一样本用户的历史购物行为和热点区域中商品的销量进行商品推荐;
    其中所述多个第一样本用户包括待推荐用户。
  3. 如权利要求2所述的商品推荐设备,其中热点区域获取模块包括:
    位置获取模块,配置成获取多个第二样本用户的位置坐标;
    聚类模块,配置成采用聚类算法,基于所述多个第二样本用户的位置坐标对所述多个第二样本用户进行聚类,得到至少一个第二样本用户类,
    热点区域确定模块,配置成根据所述至少一个第二样本用户类确定热点区域,其中与每一个第二样本用户类对应的热点区域的中心点坐标
    Figure PCTCN2019079433-appb-100001
    和半径r通过下式确定:
    Figure PCTCN2019079433-appb-100002
    Figure PCTCN2019079433-appb-100003
    其中,n为与该热点区域对应所述第二样本用户类中用户的个数,且(x i,y i)为与该热点区域对应所述第二样本用户类中第i个用户的位置坐标,其中i=1,...,n。
  4. 如权利要求2或3所述的商品推荐设备,其中所述商品推荐第一子模块包括:
    数据获取模块,配置成获取所述多个第一样本用户的历史购物行为数据;
    评分矩阵确定模块,配置成根据所述历史购物行为数据,确定第一样本用户-商品的评分矩阵;
    预测评分矩阵确定模块,配置成根据所述第一样本用户-商品的评分矩阵,利用基于矩阵分解的协同过滤算法确定所述第一样本用户-商品的预测评分矩阵;
    商品推荐第三子模块,配置成根据所述第一样本用户-商品的预测评分矩阵,向所述待推荐用户进行商品推荐。
  5. 一种商品推荐系统,包括商品推荐设备以及与商品推荐设备通信连接的用户终端,其中:
    用户终端被配置用于向商品推荐设备发送待推荐用户的位置信息;
    商品推荐设备被配置用于接收待推荐用户的位置信息,根据待推荐用户的位置信息进行商品推荐,以及将所推荐的商品信息发送给待推荐用户;并且
    用户终端进一步被配置用于从商品推荐设备接收并展示所推荐的商品信息。
  6. 一种商品推荐方法,包括:
    接收待推荐用户的位置信息;
    根据待推荐用户的位置信息,进行商品推荐;
    将所推荐的商品信息发送给待推荐用户。
  7. 如权利要求6所述的方法,其中所述根据待推荐用户的位置信息进行商品推荐的步骤包括:
    获取热点区域;
    根据待推荐用户的位置信息,确定所述待推用户是否在热点区域中;
    当待推荐用户不在热点区域中时,基于多个第一样本用户的历史购物行为进行商品推荐,;
    当待推荐用户在热点区域中时,基于所述多个第一样本用户的历史购物行为和热点区域中商品的销量进行商品推荐;
    其中所述多个第一样本用户包括待推荐用户。
  8. 如权利要求7所述的方法,其中所述获取热点区域的步骤包括:
    获取多个第二样本用户的位置坐标;
    采用聚类算法,基于所述多个第二样本用户的位置坐标对所述多个第二样本用户进行聚类,得到至少一个第二样本用户类,
    根据所述至少一个第二样本用户类确定热点区域,其中与每一个第二样本用户类对应的热点区域的中心点坐标
    Figure PCTCN2019079433-appb-100004
    和半径r通过下式确定:
    Figure PCTCN2019079433-appb-100005
    Figure PCTCN2019079433-appb-100006
    其中,n为与该热点区域对应所述第二样本用户类中用户的个数,且(x i,y i)为与该热点区域对应所述第二样本用户类中第i个用户的位置坐标,其中i=1,...,n。
  9. 如权利要求6或7所述的方法,其中所述基于多个第一样本用户的历史购物行为进行商品推荐的步骤包括:
    获取所述多个第一样本用户的历史购物行为数据;
    根据所述历史购物行为数据,确定第一样本用户-商品的评分矩阵;
    根据所述第一样本用户-商品的评分矩阵,利用基于矩阵分解的协同过滤算法确定所述第一样本用户-商品的预测评分矩阵;
    根据所述第一样本用户-商品的预测评分矩阵,向所述待推荐用户进行商品推荐。
  10. 如权利要求9所述的方法,其中所述多个第一样本用户为m个第一样本用户U1,...,Um,所述多个第一样本用户的历史购物行为涉及n种商品V1,...,Vn,并且所述根据所述历史购物行为数据,确定第一样本用户-商品的评分矩阵的步骤包括:
    根据所述历史购物行为数据,确定每个第一样本用户Ui对商品 V1,...,Vn的评分,其中在第一样本用户Ui的历史购物行为不涉及商品Vj的情况下,Ui对Vj的评分为空值,其中i=1,...,m,j=1,...,n;
    根据所述每个第一样本用户Ui对商品V1,...,Vn的评分,确定第一样本用户-商品的评分矩阵A mxn,其中矩阵A的元素A(i,j)表示用户Ui对商品Vj的评分。
  11. 如权利要求10所述的方法,其中所述根据所述第一样本用户-商品的评分矩阵,利用基于矩阵分解的协同过滤算法确定所述第一样本用户-商品的预测评分矩阵的步骤包括:
    基于矩阵分解的协同过滤算法,获得矩阵U m×k和V n×k,使得A≈U×V T,k<<m,n,
    通过下式确定第一样本用户-商品的预测评分矩阵A′
    A′=U×V T
    其中矩阵A′的元素A′(i,j)表示用户Ui对商品Vj的预测评分。
  12. 如权利要求11所述的方法,其中所述根据所述第一样本用户-商品的预测评分矩阵,向所述待推荐用户进行商品推荐的步骤包括:
    根据在第一样本用户-商品的预测评分矩阵A′中的待推荐用户对商品U1,...,Um的评分,向所述待推荐用户进行商品推荐。
  13. 如权利要求7所述的方法,其中所述根据待推荐用户的位置信息确定所述待推荐用户是否在热点区域中的步骤包括:
    计算待推荐用户的位置与热点区域的中心点之间的欧式距离;
    将所计算的距离与热点区域的半径进行比较:若所述距离大于半径,则用户不在热点区域中;否则,用户在热点区域中。
  14. 如权利要求6或7所述的方法,其中所述当用户在热点区域中时基于所述多个第一样本用户的历史购物行为和热点区域中商品的销量进行商品推荐的步骤包括:
    对所述热点区域中的商品按照销量从高到低进行排序,得到排序在前的N种商品,其中N为大于或者等于1的整数,
    获取所述多个第一样本用户的历史购物行为数据;
    根据所述历史购物行为数据,确定第一样本用户-商品的评分矩阵;
    针对所述第一样本用户-商品的评分矩阵,利用基于矩阵分解的协同过滤算法确定第一样本用户-商品的预测评分矩阵;
    根据所述第一样本用户-商品的预测评分矩阵以及所述热点区域中 排序在前的N种商品,进行商品推荐。
  15. 如权利要求14所述的方法,其中所述根据所述第一样本用户-商品的预测评分矩阵以及所述热点区域中排序在前的N种商品进行商品推荐的步骤包括:
    根据所述第一样本用户-商品的预测评分矩阵,获得至少一种第一推荐商品,且根据热点区域中排序在前的N种商品获得至少一种第二推荐商品;以及
    将所述第一和第二推荐商品以固定的或随机的优先顺序推荐给待推荐用户。
  16. 如权利要求9或14所述的方法,其中所述基于矩阵分解的协同过滤算法包括基于交叉最小二乘法矩阵分解的协同过滤算法或基于梯度下降法矩阵分解的协同过滤算法。
  17. 如权利要求6-16中任一项所述的方法,其中在根据待推荐用户的位置信息进行商品推荐之后且在所述将所推荐的商品信息发送给待推荐用户之前,还包括:
    获取所述待推荐用户已放入购物车的商品信息;
    在所推荐的商品信息中包含已放入购物车的商品的情况下,将已放入购物车的商品从所推荐的商品信息中删除。
  18. 如权利要求17所述的方法,其中所述获取所述待推荐用户已放入购物车的商品信息的步骤包括:利用购物车上的商品识别装置获取已放入购物车的商品的商品信息。
  19. 一种计算设备,包括:
    处理器;以及
    存储器,其上存储有计算机可读指令,当所述计算机可读指令被处理器执行时,使所述计算设备执行如权利要求6-18中任一项所述的商品推荐方法。
  20. 一种计算机可读存储介质,包括存储在其上的计算机可读指令,所述计算机可读指令在被执行时实现如权利要求6-18中任一项所述的商品推荐方法。
PCT/CN2019/079433 2018-06-29 2019-03-25 商品推荐方法和商品推荐设备 WO2020001096A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/500,656 US11449919B2 (en) 2018-06-29 2019-03-25 Commodity recommendation method and commodity recommendation device

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810699571.7A CN109697652A (zh) 2018-06-29 2018-06-29 一种商场的商品推荐方法和服务器
CN201810699571.7 2018-06-29

Publications (1)

Publication Number Publication Date
WO2020001096A1 true WO2020001096A1 (zh) 2020-01-02

Family

ID=66230051

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/079433 WO2020001096A1 (zh) 2018-06-29 2019-03-25 商品推荐方法和商品推荐设备

Country Status (3)

Country Link
US (1) US11449919B2 (zh)
CN (1) CN109697652A (zh)
WO (1) WO2020001096A1 (zh)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111368170A (zh) * 2020-02-11 2020-07-03 口碑(上海)信息技术有限公司 页面数据的巡检方法、装置及设备
CN112069229A (zh) * 2020-08-27 2020-12-11 贵州民族大学 一种面向移动轨迹大数据的最佳候车点推荐方法及系统
CN112926921A (zh) * 2021-03-18 2021-06-08 拉扎斯网络科技(上海)有限公司 基于智能柜的信息处理方法、装置及系统
CN113011787A (zh) * 2021-04-22 2021-06-22 中国平安人寿保险股份有限公司 基于数据聚类的虚拟对象分配方法及相关设备
CN113052651A (zh) * 2021-03-17 2021-06-29 广东便捷神科技股份有限公司 一种基于大数据的智能零售管理方法及系统
CN113077321A (zh) * 2021-04-21 2021-07-06 北京沃东天骏信息技术有限公司 物品推荐方法、装置、电子设备和存储介质
CN117853152A (zh) * 2024-03-07 2024-04-09 云南疆恒科技有限公司 一种基于多渠道的业务营销数据处理系统

Families Citing this family (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112150228A (zh) * 2019-06-28 2020-12-29 浙江汉朔电子科技有限公司 一种实体商店信息交互系统和信息推送方法
CN110248316A (zh) * 2019-07-04 2019-09-17 中国工商银行股份有限公司 基于室内定位的线上营销方法及装置
CN110837597B (zh) * 2019-10-17 2023-06-20 贝壳找房(北京)科技有限公司 拼租客推荐方法、装置、服务器及存储介质
CN111340565B (zh) * 2020-03-20 2023-05-02 北京爱笔科技有限公司 一种信息推荐方法、装置、设备及存储介质
CN113627418B (zh) * 2020-05-07 2023-08-25 杭州海康威视数字技术股份有限公司 数据管理方法、计算机设备及计算机可读存储介质
CN111931740B (zh) * 2020-09-29 2021-01-26 创新奇智(南京)科技有限公司 商品销量识别方法及装置、电子设备、存储介质
CN112785390B (zh) * 2021-02-02 2024-02-09 微民保险代理有限公司 推荐处理方法、装置、终端设备以及存储介质
CN113095876B (zh) * 2021-04-09 2022-10-18 青岛海信智慧生活科技股份有限公司 一种商品推荐方法及装置
CN114820142B (zh) * 2022-06-29 2022-09-16 国能(北京)商务网络有限公司 面向b端采购用户的商品信息推荐方法
CN116128600A (zh) * 2023-01-31 2023-05-16 读书郎教育科技有限公司 一种基于ai的推荐信息生成方法
CN116028719B (zh) * 2023-03-29 2023-06-20 杭州金智塔科技有限公司 对象推荐方法及装置、跨域联邦商品推荐方法及装置
CN116911951B (zh) * 2023-07-28 2024-03-08 北京数聚智连科技股份有限公司 一种电商数据分析处理方法及系统
CN116883121B (zh) * 2023-09-06 2023-11-14 深圳鼎智通讯有限公司 基于大数据分析的pos机用户推荐方法
CN117252620A (zh) * 2023-09-26 2023-12-19 深圳市凯必禾信息科技有限公司 一种营销信息推送系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103400291A (zh) * 2013-08-26 2013-11-20 百度在线网络技术(北京)有限公司 商品推荐方法、系统和服务器
CN106504037A (zh) * 2016-12-13 2017-03-15 合肥华耀广告传媒有限公司 一种基于地理位置的广告投放方法
CN106709764A (zh) * 2016-12-30 2017-05-24 成都圈圈儿科技有限公司 一种基于位置信息的推送方法及系统
CN107481114A (zh) * 2017-08-16 2017-12-15 北京京东尚科信息技术有限公司 商品推荐方法、装置、电子商务系统及存储介质

Family Cites Families (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1573459A4 (en) * 2002-09-20 2010-02-17 Sorensen Associates Inc SYSTEM AND METHOD FOR ANALYZING A COMMERCIAL ENVIRONMENT WITH STANDARDIZATION
US9117238B2 (en) * 2012-04-18 2015-08-25 Ebay Inc. Method, system, and medium for generating a mobile interface indicating traffic level for local merchants
CN103544632B (zh) * 2013-07-22 2016-09-21 杭州师范大学 一种网络商品个性化推荐方法及系统
US20150052003A1 (en) * 2013-08-19 2015-02-19 Wal-Mart Stores, Inc. Providing Personalized Item Recommendations Using Scalable Matrix Factorization With Randomness
US20150186920A1 (en) * 2013-12-31 2015-07-02 Spotluck Llc Optimizing and Distributing Discounts
CN104811899B (zh) * 2014-01-23 2019-11-26 腾讯科技(深圳)有限公司 一种用户活动范围确定方法、装置及终端设备
CN103996068B (zh) * 2014-05-08 2017-01-25 百度在线网络技术(北京)有限公司 客流分布的统计方法和装置
US9658693B2 (en) * 2014-12-19 2017-05-23 Immersion Corporation Systems and methods for haptically-enabled interactions with objects
CN105302916A (zh) * 2015-11-16 2016-02-03 北京百度网讯科技有限公司 信息推荐的方法及装置
CN105975578A (zh) * 2016-05-04 2016-09-28 北京奔影网络科技有限公司 信息推荐方法、装置及终端设备
CN106570722B (zh) * 2016-10-31 2020-10-30 华讯高科股份有限公司 一种智能推荐系统及智能推荐方法
CN107180378A (zh) * 2017-05-11 2017-09-19 北京旷视科技有限公司 商品关注度获得方法及装置
CN107657007B (zh) * 2017-09-22 2020-12-22 Oppo广东移动通信有限公司 信息推送方法、装置、终端、可读存储介质及系统

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103400291A (zh) * 2013-08-26 2013-11-20 百度在线网络技术(北京)有限公司 商品推荐方法、系统和服务器
CN106504037A (zh) * 2016-12-13 2017-03-15 合肥华耀广告传媒有限公司 一种基于地理位置的广告投放方法
CN106709764A (zh) * 2016-12-30 2017-05-24 成都圈圈儿科技有限公司 一种基于位置信息的推送方法及系统
CN107481114A (zh) * 2017-08-16 2017-12-15 北京京东尚科信息技术有限公司 商品推荐方法、装置、电子商务系统及存储介质

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111368170A (zh) * 2020-02-11 2020-07-03 口碑(上海)信息技术有限公司 页面数据的巡检方法、装置及设备
CN111368170B (zh) * 2020-02-11 2023-03-31 口碑(上海)信息技术有限公司 页面数据的巡检方法、装置及设备
CN112069229A (zh) * 2020-08-27 2020-12-11 贵州民族大学 一种面向移动轨迹大数据的最佳候车点推荐方法及系统
CN112069229B (zh) * 2020-08-27 2024-02-27 贵州民族大学 一种面向移动轨迹大数据的最佳候车点推荐方法及系统
CN113052651A (zh) * 2021-03-17 2021-06-29 广东便捷神科技股份有限公司 一种基于大数据的智能零售管理方法及系统
CN113052651B (zh) * 2021-03-17 2024-04-23 广东便捷神科技股份有限公司 一种基于大数据的智能零售管理方法及系统
CN112926921A (zh) * 2021-03-18 2021-06-08 拉扎斯网络科技(上海)有限公司 基于智能柜的信息处理方法、装置及系统
CN113077321A (zh) * 2021-04-21 2021-07-06 北京沃东天骏信息技术有限公司 物品推荐方法、装置、电子设备和存储介质
CN113011787A (zh) * 2021-04-22 2021-06-22 中国平安人寿保险股份有限公司 基于数据聚类的虚拟对象分配方法及相关设备
CN113011787B (zh) * 2021-04-22 2023-07-21 中国平安人寿保险股份有限公司 基于数据聚类的虚拟对象分配方法及相关设备
CN117853152A (zh) * 2024-03-07 2024-04-09 云南疆恒科技有限公司 一种基于多渠道的业务营销数据处理系统
CN117853152B (zh) * 2024-03-07 2024-05-17 云南疆恒科技有限公司 一种基于多渠道的业务营销数据处理系统

Also Published As

Publication number Publication date
CN109697652A (zh) 2019-04-30
US11449919B2 (en) 2022-09-20
US20210334875A1 (en) 2021-10-28

Similar Documents

Publication Publication Date Title
WO2020001096A1 (zh) 商品推荐方法和商品推荐设备
US20200342550A1 (en) Methods and systems for generating restaurant recommendations
US9811853B2 (en) Method and system for user based routing
US20140279263A1 (en) Systems and methods for providing product recommendations
KR102204924B1 (ko) 쇼핑 트립 플래너
US11727458B2 (en) Produce comparison system
KR102000076B1 (ko) 온라인 쇼핑몰 통합 관리 시스템에서의 온라인 판매채널을 추천하는 방법 및 그 서버
US20140304075A1 (en) Methods and systems for transmitting live coupons
US20210073892A1 (en) Product ordering method
US20190043071A1 (en) Method and system for selection, filtering or presentation of available sales outlets in a distributed networked computing environment using spatial and geography based metrics
US20220108374A1 (en) Smart Basket for Online Shopping
JP5831204B2 (ja) 情報提供システム、情報提供方法及びプログラム
CN110750697B (zh) 商户分类方法、装置、设备及存储介质
KR20150138310A (ko) 디지털 영수증 경제
US20240013281A1 (en) Systems and methods of product identification within an image
KR101639656B1 (ko) 광고 제공 방법 및 서버 장치
CN113077317A (zh) 基于用户数据的物品推荐方法、装置、设备及存储介质
JP7127096B2 (ja) 情報処理装置、情報処理方法及び情報処理プログラム
CN117421355A (zh) 一种搜索召回方法、装置及设备
KR101637986B1 (ko) 최적 상품 제공 시스템 및 그 방법
CN108416611B (zh) 一种超市路径推荐系统及其方法
KR20220156427A (ko) 인벤토리 아이템 예측 및 리스팅 추천
JP6686208B1 (ja) 情報処理装置、情報処理方法、及びプログラム
JP2021047551A (ja) 特定装置、特定方法及び特定プログラム
US20220318878A1 (en) Digital preferences based on physical store patterns

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19825857

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19825857

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 19825857

Country of ref document: EP

Kind code of ref document: A1