JP2002279279A - Product recommendation device, product recommendation method, and product recommendation program - Google Patents

Abstract

(57) [Summary] [Problem] To provide a product recommendation device, a product recommendation method, and a product recommendation program that can appropriately recommend a product to a customer in consideration of a chronological purchase order of the product. SOLUTION: Purchase history data of each customer is acquired by receiving a product order from a customer via a network and selling the product. By generating an adjacency matrix from the purchase history data and analyzing the adjacency matrix, a time-sequential order relationship between commodities related to purchase is acquired. The activity of each product is determined from the order relation, and the degree of recommendation of each product is determined from the activity and the purchase frequency of each product. When a customer purchases a product, the customer recommends to the customer a product that has a high degree of recommendation and that has a chronological order relationship with the product.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a product recommendation device, a product recommendation method, and a product recommendation program.
The present invention relates to a system for recommending a product to a user via a network.

[0002]

2. Description of the Related Art With the rapid spread of the Internet this year, the sale and recommendation of goods via the Internet have become active. There are shops established as homepages on the Internet, and shopping streets that collect these shops. The customer accesses these shops on the Internet from a terminal device such as a personal computer (hereinafter referred to as PC) or the like to search for or order a product. 2. Description of the Related Art When a customer orders a product, recommending and recommending a product that the customer is likely to order has been performed in various forms.

[0003] Such recommendation methods include, for example, a check sheet method, a rule-based method, an emphasis filtering method, and an AI method. What is the check sheet method?
Information for specifying product information that the customer wants to inquire is registered in advance, and product information that matches the registered information is provided to the customer. What is the rule-based method?
This is to guess merchandise information that the customer wants to inquire from registered personal information of the customer and provide the customer with the information.

[0004] There are two types of emphasis filtering systems, one using a product evaluation value and the other using an access history. The method using the product evaluation value is to refer to the product evaluation value of another customer to estimate the taste of the customer, and to provide products and information that the customer can refer to. The one using the access history is for referring to a product order record or an inquiry record of another customer, inferring the customer's preference, and providing product information that the customer wants to refer to. The AI (Artificial Intelligence) method is to learn customer's preference from the content of the content accessed by the customer, and to provide product information that the customer wants to inquire from the learned preference content.

[0005]

However, the check sheet method has a problem in that the customer must input in advance the merchandise information that the customer wants to search, which places a burden on the customer. Further, when the customer's preference changes, the customer must change the setting contents of the system. The rule-based method requires customers to register personal information,
When the registration load is placed on the customer and the customer changes his / her address, name, etc., the customer must change the registration. In addition, since personal information is registered in the system, there is a security problem.

[0006] In the enhancement filtering method, in the case of using the product evaluation, the customer must evaluate the product and register the evaluation result in a system, which places a burden on the customer. In the case of using the access history, if the access history of the customer is small and the system has not sufficiently learned the customer's preference, it is impossible to recommend an appropriate product to the customer. When recommending a new product to a customer that has not been accessed by the customer, the system administrator needs to set up the system so that the new product is presented to the customer. AI
In the case of the method as well, when the number of contents accessed by the customer is small similarly to the above-described emphasis filtering method using the access history, an appropriate product cannot be recommended to the customer. When recommending a new product to a customer that has not been accessed by the customer, the system administrator needs to set up the system so that the new product is presented to the customer.

[0007] Further, the above-mentioned recommendation methods merely recommend a product that matches the taste of the customer. For example, if the customer buys a television, the probability that the customer next buys a VCR increases. No recommendation has been made in consideration of the interrelationship of the chronological purchase order between the products.
For this reason, there are cases where unnecessary or meaningless recommendations are made, so that the customer cannot obtain desired product information, and the seller misses a sales opportunity. Therefore, an object of the present invention is to provide a product recommendation device, a product recommendation method, and a product recommendation program that can appropriately recommend a product in consideration of a chronological purchase order of the product.

[0008]

According to the first aspect of the present invention, a product information obtaining unit for obtaining product information on a product from a user, and a specifying unit for specifying a product from the product information obtained by the product information obtaining unit. And a product estimating means for estimating a specific product specified by the specifying device, a related product having a chronological order relationship with respect to purchase, and an expectation that the user will purchase the relevant product after purchasing the specific product. A degree of expectation acquisition unit for acquiring the estimated degree of each related product, and, among the related products, product information of a related product having a high degree of expectation acquired by the expectation degree acquisition unit with priority to the user. The above object is achieved by providing a product recommendation device with a recommendation means for recommending a product. According to a second aspect of the present invention, in the product recommendation apparatus according to the first aspect, a history recording unit that records a purchase history of the product, and a purchase history of the product recorded by the history recording unit. A time-series relation estimating means for estimating a time-series order relation regarding purchase between the user and the time-series purchase order between the products estimated by the time-series relation estimating means, and the user identifies the specific product. Further comprising a possibility degree obtaining means for estimating a degree of purchase possibility of purchasing the related product after the purchase,
The expectation obtaining unit obtains the expectation using the purchase possibility obtained by the possibility obtaining unit. According to a third aspect of the present invention, in the product recommendation apparatus according to the second aspect, the time-series relation estimating means includes an adjacency matrix that can specify a time-series order in which a user purchases a product by a position of a matrix element. Further comprising a matrix generation means for generating, wherein the time-series relation estimating means specifies a related product having a time-sequential order relationship with the specific product from a position of a matrix element of the matrix generated by the matrix generation means. It is characterized by doing. According to a fourth aspect of the present invention, in the commodity recommendation device according to the second or third aspect, the adjacent matrix generated by the matrix generating unit further includes a value of the matrix element for the adjacency matrix. After the user purchases a specific product specified from the position, the user purchases a related product specified from the position of the matrix element, and the likelihood degree obtaining unit indicates the degree of possibility of the matrix element of the adjacent matrix. After the user purchases the related product using the value, a degree of purchase possibility of purchasing another product is acquired. According to a fifth aspect of the present invention, in the product recommendation apparatus according to the second, third or fourth aspect, a purchase frequency for acquiring a purchase frequency of each product from a purchase history recorded by the recording means. Acquisition means, wherein the expectation degree acquisition means uses the purchase frequency acquired by the purchase frequency acquisition means and the degree of purchase possibility acquired by the possibility degree acquisition means to obtain each of the associations. It is characterized by acquiring the degree of expectation of a product. Claim 6
According to the invention described in Item 1, in the product recommendation device according to any one of Claims 1 to 5, the product information obtaining means obtains product information from the user via a network. The recommendation means recommends product information to the user via the network. In the invention according to claim 7, a first step of obtaining product information on a product from a user, a second step of specifying a product from the product information obtained in the first step, and The third step of estimating the specific product specified in the step and the related product having a chronological order relationship with respect to purchase, and the degree of expectation that the user purchases the related product after purchasing the specific product, A fourth step of acquiring for each related product estimated in the third step;
A fifth step of preferentially recommending to the user the product information of the related product having a high degree of expectation acquired in the step (c). In the invention according to claim 8, a product information obtaining function for obtaining product information on a product from a user, a specifying function for specifying a product from the product information obtained by the product information obtaining function, and a specifying function for specifying the product. The specified product, and a product guessing function for guessing a related product having a chronological order relationship with respect to purchase, and the degree of expectation that the user purchases the related product after purchasing the specific product,
The expected degree acquisition function to be acquired for each of the estimated related products, and, among the related products, the product information of the related product having a high degree of expectation acquired by the expected degree acquisition function is preferentially recommended to the user. The object is achieved by a product recommendation program for realizing a recommendation function and a computer. It is also possible to provide a computer-readable storage medium on which the product recommendation program described in claim 8 is recorded.

[0009]

BEST MODE FOR CARRYING OUT THE INVENTION (1) Overview of the embodiment There are many products that have a chronological order relationship regarding purchase between products. Some chronological order relationships are self-evident from the characteristics of the product itself, while others can only be revealed by actually having many customers purchase the product and statistically processing those purchase patterns. . By recommending products in consideration of such a chronological purchase pattern (a chronological order relation regarding product purchase), it is possible to continually link many products to sales triggered by the purchase or recommendation of one product. Can be. As a result, sales can be increased. Therefore, in the present embodiment, an adjacency matrix that reflects the chronological order relationship and purchase frequency of purchased products for each customer is obtained from the product purchase history data, and the sum of this adjacency matrix is obtained for all customers to obtain It is a chronological purchase pattern. This becomes the statistical basic data of the chronological order relation regarding the product purchase. Based on this order relationship,
When a certain product is purchased, a product having a high probability of being subsequently purchased is recommended.

In the present embodiment, when a customer orders or inquires for a product via a network, the product that is temporally related to the purchase of the product is presented to the customer and effective product recommendation is performed. Is what you do. Here, the product having a chronological order relationship with the purchase of a certain product is defined as a starting point of the purchase of the product a, for example, a customer purchases a printer and then purchases toner and printing paper for the printer. Means a relationship in which another product b is purchased. At this time, by recommending (recommending) a product as a starting point in the chronological order as much as possible to the customer, the customer can know the related product of the purchased product, while the seller (store) side has more We can expect customers to purchase many products.

Hereinafter, a preferred embodiment of the present invention will be described in detail with reference to FIGS. Figure 1
FIG. 3 is a diagram showing a relationship between a product recommendation device and a user terminal device according to the present embodiment. The product recommendation device 2 is connected to the user terminal devices 4, 4, 4,... There are a plurality of user terminal devices 4, 4, 4,.
Hereinafter, description will be made using one of the user terminal devices 4.

The product recommendation device 2 is, for example, a server that provides product information to a customer (hereinafter referred to as a user) and receives an order or an inquiry for a product from a user. Then, the user can order or refer to the product recommendation device 2 for a product from the user terminal device 4 via the network 3. In the product recommendation device 2, one or more product distributors open a homepage, thereby opening a virtual store. When the user visits a store opened in the product recommendation device 2 from the user terminal device 4 and makes an order or inquires for the product, the product recommendation device 2 outputs the products in a chronological order relationship of purchase described later. Is displayed on the user terminal device 4 and recommended.

In the present embodiment, the network 3 is assumed to be the Internet. However, the present invention is not limited to this, and a LAN (Local Area Network) is used.
k), WAN (Wide Area Network)
k), a network utilizing an artificial satellite network or a digital television broadcast two-way communication is also included in the scope disclosed in the present embodiment. The term “product” in the present embodiment is not limited to a product such as an electric product, but may be a product such as a financial product or an insurance product that does not have a physical substance, a service such as transportation and painting, and event information. It shall also include information provision.

FIG. 2 is a diagram showing an example of the internal configuration of the product recommendation device 2. In the product recommendation device 2, the central processing unit 5
7, a storage unit 61, an input / output unit 58, a communication control unit 59, and the like are connected by a bus line 60. The central processing unit 57 controls each component of the product recommendation device 2 according to a predetermined program and performs a computation (C
RO, which is a read-only memory in which basic programs and various parameters for operating the product recommendation device 2 are stored.
RAM which is a readable / writable memory used when providing a working memory to the M (Read Only Memory) and the CPU and when loading various programs
(Random Access Memory).

The storage section 61 stores a product recommendation program described later, other various programs 62, and various database sections 63. The product recommendation device 2
By loading the commodity recommendation program stored in the program unit 62 into the central control unit 7 and executing it, the apparatus can function as an apparatus having a commodity recommendation function described later.

The input / output unit 58 includes, for example, a display,
It is connected to a keyboard, a mouse, a floppy (registered trademark) disk drive, a CD-ROM drive, and the like, and a user interface between the administrator of the product recommendation device 2 and the product recommendation device 2 and data from outside the product recommendation device 2. It provides an interface for communication. The communication control unit 59 is, for example, a modem or a terminal adapter, and the product recommendation device 2 is connected to the network 3 via the communication control unit 59. The communication control unit 59 controls communication between the product recommendation device 2 and the network 3. The bus line 60 transmits data input / output from the input / output unit 58, the communication control unit 59, the storage unit 61, and the like to a destination based on the timing of a clock generated by the central control unit 7.

FIG. 3 is a diagram showing a configuration of a product recommendation system 15 configured by using the product recommendation device 2. The product recommendation system 15 is realized in software by loading the product recommendation program stored in the program unit 62 of the storage unit 61 into the central processing unit 57 and executing the product recommendation program in the product recommendation device 2. Further, each component may be configured by hardware. Also,
The product recommendation device 2 includes, in addition to the product recommendation system 15, a product order receiving system (not shown) that receives and processes an order for a product from a user.

The product recommendation system 15 is a system for recommending and advertising an appropriate product to the user when receiving an order or inquiry for a product from the user terminal device 4. Hereinafter, the configuration of the product recommendation system 15 will be described. The product recommendation recommendation system 15 includes an order receiving unit 7 and a purchase history database 6. When a user places an order for a product, the central control unit 57 of the product recommendation device 2 displays a product order screen on the display of the user terminal device 4, and the user displays a product code, a quantity, and the like on the product order screen. By inputting, an order for a product can be made. In the present embodiment, the product order screen is prepared for each store when the customer visits each store on the network. It should be noted that this can also be configured so that orders at each store are received collectively on a single order screen.

The order receiving unit 7 is a component that receives order information input from the user terminal device 4 from the product order screen via the network 3 and performs appropriate processing described later. The order receiving unit 7 is a user terminal device 4
The order information acquired from is, for example, a user name, a user code, a product code, an order quantity, an order date and time, and the like.

The purchase history database 6 includes an order receiving unit 7
Receive order information from and record it as purchase history. In the purchase history database 6, order information obtained from each user is recorded retroactively. The purchase history database 6 is provided in the database unit 63 of FIG. From the purchase history recorded in the purchase history database 6, it is possible to know the chronological order relation of the products purchased by each user (the temporal order of purchase between the products).
Then, by statistically analyzing the temporal purchase order together with information such as the occupation of the user, for example, a certain user tends to purchase the product b at a certain time interval after purchasing the product a. It is possible to know the strong chronological order of product purchase from various viewpoints such as each user, gender, and occupation.

The adjacency matrix generation unit 5 acquires the purchase history of each user from the purchase history database 6 and generates an adjacency matrix for each user later, and further adds an adjacency matrix generated for each user. It is a component that generates the entire adjacency matrix. The adjacency matrix is a matrix generated from the order in which the user purchased the products, and expresses the purchase order relationship between the products by the elements of the matrix. Details will be described later.

The user information database 12 stores the age, gender, occupation, hobbies,
It is a database in which user information such as an address and a telephone number is recorded, and is a component that searches these data and provides the data to the adjacency matrix analysis unit 8, the recommended information search unit 11, and the like. The user information database 12 is provided in the database unit 63 of FIG.

The adjacency matrix analysis unit 8 analyzes the adjacency matrix generated by the adjacency matrix generation unit 5 from various angles such as gender, age,... Then, when a certain user places an order for a product, the recommendation information database 1 determines which product is most efficient to recommend and advertise to the user according to a pre-programmed algorithm.
3 is a component to be output. Adjacency matrix analyzer 8
Comprises a purchase frequency calculation unit 9, an activity calculation unit 10, and a recommendation degree calculation unit 16.

The purchase frequency calculation unit 9 is a component for counting the purchase frequency for each product. The activity calculation unit 9
It is a component that calculates the activity of each product, and the activity of a product is an index indicating how much other products are continuously purchased due to the user purchasing the product. For example, the activity level of the product a means that when the user purchases the product a, the purchase of the product a starts from, and then the user purchases a large number of products such as the product b, the product c,. It is a quantitative expression of the degree of cause. An example of a specific definition using a mathematical expression will be described later. The recommendation degree calculation unit 16 includes the purchase frequency calculation unit 9
Is a component for quantitatively calculating the recommendation level for each product from the purchase frequency totalized in the above and the activity level calculated by the activity level calculation unit 10. An example of a specific calculation formula of the recommendation degree will be described later.

The recommendation information database 13 includes, for each product, information indicating a degree of recommendation and a product having a chronological order with the product, information specifying a position of the product in the chronological order, The activity and the like of the product are stored separately for each user's gender, age, occupation, residence area, and the like. The data recorded in the advertisement database 13 is configured so that when a certain user orders a certain product, the attribute of the user and the product recommended to the user can be specified.

For example, if the ordered product is a canned set called “Hokkaido Taste Set” and the attribute of the user who ordered is “woman” or “20 years old”, the recommended product is “Hokkaido Ski Tour” planned by a travel agency. If the attribute of the user is “female” or “over 60”, the recommended products are “trip of Michinoku Onsen” planned by the travel agency and “kitchen parcel set” provided by the food company. It has become. In the above example, gender and age are used as user attributes. However, the present invention is not limited to this, and it is possible to analyze various combinations of user attributes.

For example, the recommendation information database 13 can be created so as to associate ordered products with recommended products for each user. For example, from the past purchase record of the specific user X recorded in the purchase history database 6, it can be seen from the analysis result of the adjacent matrix analysis unit 8 that the user X tends to purchase the product b after purchasing the product a. In this case, the adjacency matrix analyzer 8 recommends the recommendation information database 13 so as to recommend the product b when the user X orders the product a.
Create

The recommendation information retrieval unit 11 acquires a product ID for specifying a product ordered from the order receiving unit 7 and a user ID for specifying a user who has ordered. Then, by using the user ID, the attribute of the user is searched and acquired from the user information database 12. Next, a product recommended to the user is searched from the recommendation information database 13 using the attribute of the user and the product ID. For example, it is assumed that the ordered product is a canned set called "Hokkaido taste set" and the attribute of the user who ordered is "female" and "20s". At this time, the recommendation information search unit 11 sends the recommendation information database 1
From 3, search for "Hokkaido Ski Tour" as a recommended product. The recommendation information search unit 11 sends the search result to the visualization unit 14.

The visualization unit 14 is a component that visualizes information of a product recommended by the user from the recommended information search unit 11 and displays the information on the display of the user terminal device 4 via the network 3. As will be described later, the visualization information unit 14 can visually represent the product searched by the recommendation information search unit 11 together with the activity.

In the commodity recommendation system 15 configured as described above, the order receiving unit 7, the recommendation information searching unit 11, the visualizing unit 14, and the purchase history database 6 are provided in real time every time an order is received from the user terminal device 4. It works on. That is,
When the user places an order for a product from the user terminal device 4 to the order receiving unit 7, the order log file in the purchase history database 6 is immediately updated, and the recommendation information search unit 1 is updated.
The product recommended by the user 1 is searched, and the recommendation information of the product is displayed on the user terminal device 4 by the visualization unit 14. On the other hand, the adjacency matrix generation unit 5 and the adjacency matrix analysis unit 8 are started, for example, once a day by a scheduler (not shown),
The recommendation information database 13 is updated.

FIG. 4 is a diagram for explaining an adjacency matrix. The adjacency matrix 21 is generated by the adjacency matrix generation unit 5 for each user from the purchase records of the users recorded in the purchase history database 6 of FIG. The rows and columns of the adjacency matrix 21 are products x, y, z,... Arranged in an appropriate order, respectively, and the adjacency matrix 21 is a square matrix. The user is x
Suppose that the goods are purchased in the order of → y → z → y. Then, the adjacency matrix generator 5 generates (x, y), (y,
z), (z, y) elements, ie, matrix element 22, matrix element 2
3. Add 1 to the matrix element 24.

That is, when the user purchases a product b next to the purchase of the product a, 1 is added to the matrix element whose row is the product a and whose column is the product b. As described above, the adjacency matrix 21 reflects the chronological order relation of the product purchase and the purchase frequency. It should be noted that the merchandise whose purchase order is obvious is inevitably purchased by the user as well, so that the adjacency matrix 21 is naturally reflected in the matrix elements by totaling, so there is no need to consider it.

In this embodiment, the adjacency matrix is defined as described above. However, the present invention is not limited to this. For example, as defined by the following equation (1), after the product a is purchased, It is also possible to define a matrix element that reflects the time required until the purchased product “b” is purchased, and add the matrix element each time the user places an order. k / (Δt + c) (1) Here, Δt is a time elapsed from when the user purchases the product a to when the user purchases the product b, and k and c are expressed by Expression (1). Is a constant for normalizing the value to be 0 or more and 1 or less. As can be seen from the fact that Δt is in the denominator, the value represented by Expression (1) increases as the time difference from when the user purchases the product a to when the user purchases the product b is smaller.

As described above, the method of adding 1 (or the purchased number or quantity) to the matrix element or the method of adding the value represented by the formula (1) is merely an example. The weights are defined for each item, and when the product is purchased, the predetermined matrix elements are operated (addition, subtraction, multiplication, division, division, Calculation such as square root, logarithm, or the like, or calculation using a predetermined function).

As the user repeatedly purchases the product,
Since the matrix elements of the adjacency matrix 21 become large, by looking at the products corresponding to the matrix elements, it is possible to infer the chronological order relation between the products related to purchase and the purchase frequency. Since the adjacency matrix 21 is created for each user, depending on the product, the adjacency matrix 21 may depend on attributes such as the user's preference, age, and gender, or may be a general one that applies to all users. There is also.

The adjacency matrix generator 5 shown in FIG. 3 generates adjacency matrices that are further statistically processed from the adjacency matrices 21, 21, 21,... Obtained for each user. For example, the adjacency matrix generation unit 5 generates the adjacency matrices 21, 21, 2 of all the users.
By adding 1,..., An overall basic adjacency matrix for expressing a chronological order relation regarding product purchase is generated.
Or, for example, classify users into several classes according to user attributes such as "employee" and "male"
The adjacency matrix of each user is added in each class to determine the temporal order of product purchase for each class. Also, the adjacency matrix 21
For example, when asked for each product category such as car-related products, audio-related products, digital device-related products,
The accuracy of the adjacency matrix 21 for each category can be improved.

FIG. 5 shows the adjacency matrix 21 visualized using a graph structure. Thus, any adjacency matrix can be visualized using a graph structure. When expressing the adjacency matrix with a graph structure, the product name is a node,
The size of the matrix element is represented by the line width of the edge (arrow), and the order of purchase is represented by the direction of the edge. However, when the matrix element is 0, the edge width is 0. In FIG.
The larger the value of the matrix element, the wider the edge width. That is, the wider the width of the edge, the greater the possibility that the user purchases the product at the end of the edge after purchasing the product at the start of the edge.

For example, when there are three types of products, x, y, and z, as shown in FIG. 5, the respective products correspond to the nodes 26, 27, and 28. The purchase order in which the user purchases the product x after purchasing the product x is represented by the direction of the edge 29 from the node 26 to the node 28, and the size of the matrix element when the user purchases the product x and then purchases the product z Is represented by the thickness of the edge 29.

Similarly, the relationship between the product x and the product y in the adjacency matrix 21 is visualized by the nodes 26, 27 and the edge 30, and the relationship between the product y and the product z is represented by the node 27,
Visualized by nodes 28 and edges 31. FIG.
In the example of, since the width of the edges is wide in the order of the edges 31, 30, and 29, if the user purchases the product y, then the user is most likely to purchase the product z, and the product x
It is found that the probability of purchasing the product y after the purchase of the item is the next highest, and the probability of purchasing the product z after the purchase of the product x is lowest.

Hereinafter, an example of a method of calculating the activity of a product performed by the activity calculator 10 of FIG. 3 and a method of calculating the recommendation performed by the recommendation calculator 16 will be described. The purchase frequency calculation unit 9 simply counts the frequency at which the product is purchased (the number of times the user purchases the product during a unit period), and a description thereof will be omitted.

(Method of Calculating Product Activity) A basic form of product activity is defined by the following equation (1). (Activity of product a) = Σv (a, j) −Σv (i, a) (1) where v represents an adjacency matrix, and v (a, j), v (i, a)
a) represents a matrix element (a, j) and a matrix element (i, a) of the adjacency matrix v, respectively. Further, the first term on the right side of (1) is the sum for j, and the second term is the sum for i. The first term indicates the degree at which the product a is a starting point for purchasing another product, and the second term indicates the degree at which the product a is purchased as a result of the user purchasing the other product. The higher the activity defined in (1) is, the more the product is located on the starting side in the chronological order of purchase, and the lower the activity is, the closer to the end in the chronological order of purchase. It is a product located in. In other words, the activity indicates the degree of expectation that the user can expect to purchase more products due to the purchase of the product after purchasing the product for each product.
In addition, a more accurate expectation can be obtained by comprehensively determining the activity and the purchase frequency. In the present embodiment, the higher the degree of recommendation, the higher the degree of recommendation. The expression (1) does not limit the definition of the activity. For example, as shown in the following expression (11), it is defined by the first term of the expression (1), or As shown in (12), a ratio obtained by dividing the first term of the equation (1) by the second term (in this case, it is like a ratio of an output to an input, which means a so-called amplification factor) may be used. . (Activity of product a) = ａv (a, j) (11) (Activity of product a) = Σv (a, j) / Σv (i, a) (12)

FIGS. 6 (a) and 6 (b) are diagrams for explaining the definition of the activity of a product according to equation (1). Now, it is assumed that there is a purchase time relationship between the product c and the product d, and the user purchases the product c and then purchases the product d. At this time, the order relationship between the purchase of the product c and the product d is represented by an edge having the product c as a start point and the product d as an end point. FIG.
In the figure, the product nodes around each edge are omitted. That is, around the product a, the products a1 to a10
However, it is assumed that there are products b1 to b11 around the product b, but these products are not shown. or,
For the sake of simplicity, it is assumed that the purchase frequency from each product is the same (once), and the line width of each edge is the same.

FIG. 6A shows an edge starting from or ending with the product a. Other products that are in chronological order with the product a are not shown. The number of edges starting from product a (facing outward from product a) corresponds to the first term of equation (1), and the number of edges starting from product a (facing product a). Corresponds to the second term of equation (1). In this example, the number of edges at which the product a is the starting point is eight, and the number of edges at which the product a is the ending point is two.
Has an activity of 8-2 = 6. Similarly, the activity of the product b in FIG. 6B is 2-9 = -7. By calculating the activity in this way, it can be seen that the product a is the starting product and the product b is the ending product in the chronological order of purchase.

That is, when the user purchases the product a, there is a high possibility that the user will subsequently purchase another product, and when the user purchases the product b, the user will purchase another product subsequently. The small possibility can be quantitatively grasped by calculating the activity.

For example, when the user purchases a personal computer (hereinafter, referred to as a PC), the purchase of the PC prompts the purchase of other peripheral products such as an expansion memory, a printer, a modem, and the like. Prompt you for the purchase of printing paper, and the modem prompts you to join a telephone company or Internet provider and purchase a telephone line cable. As described above, PC is considered to be a product having extremely high activity. On the other hand, the above toner is considered to be a product having a low activity, which is close to the end product.

The product which is the starting point of the purchase of a series of products is a starting (initiator) product, and the ending product is an end (terminator) product. The activity of a product quantitatively represents how close the product is to the starting product or to the end product. As in the example of the PC above, when there is a temporal order relationship of purchase between certain products purchased by the user, the younger the order is, the more active the product is to encourage the user to purchase another product. Become.

A purchase pattern useful in product recommendation is that the purchase frequency of the product itself is high, and that the number of products purchased or the sales price is large following the purchase of the product. A product satisfying these two conditions is desirable as a recommendation candidate.

Although the basic form of the activity of the product a is defined by the equation (1), the activity is defined for a more complicated case. [Case where a plurality of products are purchased at the same time] FIG.
(Commodity b, commodity c,...) Are purchased at the same time, and thereafter, a total of M commodity groups (commodity 1, commodity m,...) Are purchased at the same time, and then the commodity X is purchased. As described above, when the user purchases a plurality of products at the same time, it is not possible to define a chronological order relationship of purchase between these products. In such a case, as an example, it is defined that an evenly divided numerical value is added to the corresponding element of the adjacent matrix.

By defining the above activities, the matrix elements of the adjacency matrix are calculated as follows. v (a, b) = v (a, b) + 1 / N (2) v (a, c) = v (a, c) + 1 / N ... v (b, l) = v (b, l) + / NM v (b, m) = v (b, m) + 1 / NM ... v (l, X) = v (l, X) + 1 / M v (m, X) = v (m, X) +1 / M... Here, for example, as shown in Expression (2), X = X + Y
The expression of the form means that a value obtained by adding Y to X is used as a new value of X. That is, 1 is added to the matrix element v (a, b).
This means that the value obtained by adding / N is used as a new matrix element v (a, b).

[When a product is bought as a set with another product] As an example of such a product, for example, a set of coffee (comprising a coffee bowl, a saucer, a sugar pot, a milk container, a spoon, etc.) And furniture consisting of a set of chairs and a table. Note that these products are usually purchased at the same time, but the present invention is not limited to this, and may be products that form a set and are purchased in chronological order. In such a case, if a graph of these products is drawn, it becomes as shown in FIG. 8A. When one of the products forming the set is purchased, the other products forming the set are subsequently purchased. Become. In this case, the graph structure of the products a, b, c, and d is a complete graph in which each edge has almost the same width and all nodes are connected.

As shown in FIG. 8B, with respect to commodity groups constituting such a complete graph, scores of all edges entering and exiting the complete graph are aggregated as shown in FIG. The degree of activity for is determined. Further, when any one of the commodities constituting the complete graph is purchased, other commodities are also purchased. That is, for example, it is assumed that there are N products composing the complete graph. Then, when the user purchases any one of the commodities constituting the complete graph, the user also purchases other N-1 commodities. for that reason,
The activity of each product constituting the complete graph can be defined by the following equation (3). (Activity of each product constituting the complete graph) = (Activity of complete graph) + N-1 (3) According to the above definition, the activities of the products constituting the complete graph are equal to each other.

[Case where the time-series order of purchase forms a loop] In the case of a product forming a self-loop (a product that is purchased periodically like a detergent), the purchase frequency calculation unit 9 grasps it. No need to consider. Otherwise, FIG.
As shown in FIG. 9A, with respect to commodity groups forming a closed loop in which the values of each edge are substantially equal, these are regarded as one block, and the activity of the entire closed loop is reduced as shown in FIG. 9B. Ask. If any one of the commodities constituting the closed loop is purchased, other commodities are purchased. Therefore, similarly to the case where the above commodities are purchased as a set, the activity of each composing product constituting the closed loop is It can be defined by the following equation (4). (Activity of each product composing the closed loop) = (activity of the closed loop) + N−1 (4) According to the above definition, the activity of each product composing the closed loop is equal to each other.

(Method of Calculating Product Degree of Recommendation) As described above, the product activity is defined. By determining the purchase frequency and activity of the product comprehensively, more effective product recommendation ( Advertising). In other words, the higher the purchase frequency of the product itself and the higher the activity, the more desirable the recommended product. In order to quantitatively express such a recommendation degree, as an example, the recommendation degree R (a) of the product a is defined as in the following equation (5). R (a) = (Wh × ah + Wp × ap) / (Wh + Wp) (5)

Here, ah and ap are the purchase frequency and activity, respectively, normalized by the following equations (6) and (7) and converted to values from 0 to 1, and Wh and Wp are appropriate weights. It is a coefficient. ah = (purchase frequency of product a) / (maximum purchase frequency) (6) ap = (activity of product a−minimum activity) / (maximum activity−minimum activity) (7) The purchase frequency is the purchase frequency of the product with the highest purchase frequency among the products constituting the adjacency matrix, and the maximum activity and the minimum activity are the maximum value and the minimum value of the matrix elements of the adjacency matrix, respectively. .

When a user places an order for a product, the recommendation information search unit 11 shown in FIG.
3, a product having a chronological order relationship with the product is searched. A product having a high degree of recommendation defined by Expression (7) is selected from the searched products, and the product selected by the visualization unit 14 is selected. Chronological order and activity. The visualization unit 14 displays the products sent from the recommendation information search unit 11 on the user terminal device 4 as a graph structure as shown in FIG.

The operation of the product recommendation device 2 configured as described above will be described. In FIG. 2, the product recommendation device 2
Displays a login screen on the user terminal device 4 via the network 3. User has password or user ID
To log in to the product recommendation device 2. Next, in the product recommendation device 2, the central processing unit 57 shown in FIG.
Display a product order screen also serving as a product guide screen of each store. The user inputs a product ID, a quantity, and the like on the product order screen, and transmits the input data to the order receiving unit 7 shown in FIG.

The order receiving unit 7 transmits the data received from the user to the recommendation information search unit 11 and the purchase history database 6. When the purchase history database 6 receives the order information from the order receiving unit 7, the purchase history data is updated. The recommendation information search unit 11 acquires a user attribute from the user information database 12, and acquires a product to be recommended from the recommendation information database 13 based on the acquired user information and the product ID. The recommendation information search unit 11 is configured to execute the recommendation information database 1
The information of the product acquired from 3 is sent to the visualization unit 14. The visualization unit 14 graphs the product information acquired from the recommendation information search unit 11 and displays the product information on the user terminal device 4.

On the other hand, a scheduler (not shown)
For example, once a day, the recommendation information database 13 is updated as follows. The adjacency matrix generator 5 receives data from the purchase history database 6 and generates an adjacency matrix for each user. In addition, all the generated adjacency matrices are added to generate an entire adjacency matrix, or a sum of adjacency matrices is calculated for each class for users classified into classes by attributes, or for each product category. Or the sum of adjacency matrices.

The adjacency matrix analysis unit 8 stores various adjacency matrices generated by the adjacency matrix generation unit 5 and the user information database 12.
The purchase frequency and the activity level are calculated for each product from the information acquired from, and the recommendation level is calculated for each product from the calculated two types of values. The adjacency matrix analysis unit 8 updates the recommendation information database 13 with the calculated purchase frequency, activity level, recommendation level, and the like.

Next, a modification of the embodiment will be described. In the above example, when the user purchases a product, a product having a high degree of recommendation is displayed on the user terminal device 4 among products having a chronological order relationship between the product and the purchase. A more advanced product recommendation method is possible by using a function that allows the user to know in advance which products are expected to sell in the future. Hereinafter, a recommendation method showing a time-series path of product purchase will be described with reference to FIGS. 10 and 11 as an example.

FIG. 10 shows an example of a time-series path of a product displayed on the display unit of the user terminal device 4 by the central processing unit 57 of the product recommendation device 2 via the network 3 according to the visualization unit 14. In this example, a product with a high purchase probability is presented / recommended with a pass following a product purchased or interested by the user. In this example, the recommendation degree is represented by the thickness of the edge. By displaying a group of related products with a pass in this way, the user can be made aware of the relationship between products that were not noticed at the time of purchase and the necessity of other products, and can expect bulk purchases. It can lead to more product sales.

More specifically, FIG. 10 is displayed when the user purchases the book “Everyone can understand financial engineering first volume”. "Everyone can understand financial engineering
The time series related products of the first volume include “Financial Engineering Understandable for Everyone”, “Financial Engineering Understandable for Everyone”, “History of Finance”, “Macroeconomics and Financial Engineering”,
There is a “Financial Term Dictionary” (both books) and the like, and the degree of recommendation is indicated by the thickness of the edge (edge).

In the case where a product with high activity is indirectly linked to a product with high activity, it is possible to encourage the user's awareness by explicitly recommending the product without directly recommending the product with high activity. One example is shown in FIG. FIG. 11 is a graph structure displayed on the user terminal device 4 according to the operation of the visualization unit 14 configured as software in the product advertisement device 2. For example, suppose that the user purchased a book called “Hyundai Camera”. When "Hyundai Camera" is indirectly linked to "Digital Camera" which is a very active digital device product, the highly active "Digital Camera" is displayed directly from "Hyundai Camera" and is not recommended Related books such as "You are a photographer with a digital camera" and "Introductory digital camera" are displayed.

In addition, merchandise can be cooperatively recommended as follows (collaborative merchandise sales). When the user purchases a certain product, the operator of the product recommendation device 2 contacts a store that sells products likely to be subsequently purchased, and provides the store with an opportunity for product recommendation. The store performs product recommendation including, for example, a discount to the user via the network 3, and when this product recommendation activity is linked to the sale of the product, the operator of the product recommendation device 2 pays a certain success fee. Shall receive it.

Further, by extracting users having similar product purchase patterns as described below, more effective product recommendation activities can be performed. The adjacency matrix for each user generated by the adjacency matrix generation unit 5 of FIG. 3 is compared,
Extract similar users. When a similar user orders a product from a product purchase pattern of a similar user, a product to be purchased in the future is predicted, and the predicted product is displayed on the user terminal device 4 and recommended. The similarity between adjacent matrices may be determined, for example, by defining the sum of squares of the difference between the values of each matrix element of the adjacent matrix as a distance, and determining that users with smaller distances are more similar to each other.

In the above embodiment, the following effects can be obtained. It is possible to grasp the chronological order relation when the user purchases a product. Further, the activity of a product can be quantitatively calculated from the chronological order relationship, and the recommendation degree of the product can be quantitatively calculated from the activity of the product and the purchase frequency. When recommending a product to a user, the recommendation is made in consideration of the chronological order of purchase. Therefore, instead of purchasing a product and ending with the product, the user continues to purchase another product. Products with a high probability of receiving can be recommended, and more sales can be expected. Also, from the user's point of view, it is possible to notice products that were not noticed at the time of product purchase,
You can shop with high efficiency. In addition, since an adjacency matrix is generated for each user, not only can an adjacency matrix be generated for various combinations of user attributes, but also a highly accurate adjacency matrix can be created by variously combining user attributes and product categories. Can be generated.

In the present embodiment and the modified example, the adjacency matrix is obtained by using the purchase history data of the product. However, the present invention is not limited to this. For example, Web (Wor
(ldWide Web) An access log of the page may be recorded, and an adjacency matrix relating to the temporal access order of the Web page may be created from the access log. In this case, the content of the product or the like described on the accessed page is the recommended content. Further, in the present embodiment and the modified example, when the user places an order for a product, the recommended product is displayed on the user terminal device 4. For example, when the user requests detailed information on the product, Can be configured to recommend a product when inquiring about.

[0068]

According to the present invention, it is possible to provide a product recommendation apparatus, a product recommendation method, and a product recommendation program which can appropriately recommend a product in consideration of a chronological order relation of purchase of the product. it can.

[Brief description of the drawings]

FIG. 1 is a diagram showing a relationship between a product recommendation device and a user terminal device according to the present embodiment.

FIG. 2 is a diagram illustrating an example of an internal configuration of a product recommendation device.

FIG. 3 is a diagram showing a configuration of a product recommendation system.

FIG. 4 is a diagram for explaining an adjacency matrix.

FIG. 5 is a diagram in which an adjacency matrix is visualized using a graph structure.

FIG. 6 is a diagram for explaining the activity of a product.

FIG. 7 is a diagram for explaining handling of activity when a plurality of products are purchased at the same time.

FIG. 8 is a diagram for explaining the handling of the activity when a certain product is purchased as a set with another product.

FIG. 9 is a diagram for explaining the handling of the degree of activity when the chronological order of purchase forms a loop.

FIG. 10 is a diagram for describing a modification of the present embodiment.

FIG. 11 is a diagram for explaining another modification of the present embodiment.

[Explanation of symbols]

 2 Product recommendation device 3 Network 4 User terminal device 5 Adjacency matrix generation unit 6 Purchase history unit 7 Order reception unit 8 Adjacency matrix analysis unit 9 Purchase frequency calculation unit 10 Activity calculation unit 11 Recommendation information search unit 12 User information database 13 Recommendation information Database 14 Visualization unit 15 Product recommendation system 16 Recommendation degree calculation unit 21 Adjacency matrix 22 Matrix element 23 Matrix element 24 Matrix element 26 Node 27 Node 28 Node 29 Edge 30 Edge 31 Edge 32 Commodity group 33 Commodity group 57 Central processing unit 58 Enter Output unit 59 Communication control unit 60 Bus line 61 Storage unit 62 Program unit 63 Database unit

Claims (8)

[Claims] A product information acquiring unit that acquires product information on a product from a user; a specifying unit that specifies a product from the product information obtained by the product information obtaining unit; and a specific product that is specified by the specifying unit And product estimating means for estimating a related product having a chronological order relationship with respect to purchase; and acquiring the expected degree of purchasing the related product after the user has purchased the specific product for each of the estimated related products. And a recommendation unit that preferentially recommends to the user the product information of the related product having a high degree of expectation acquired by the expectation acquisition unit among the related products. Characteristic product recommendation device. 2. A time series for estimating a chronological order relation between products by using a history recording means for recording a purchase history of the goods, and a purchase history of the goods recorded by the history recording means. Relationship estimating means, using the time-series purchase order between the products estimated by the time-series relationship estimating means, the degree of purchase possibility that the user purchases the related product after purchasing the specific product And a probability degree acquiring means for estimating the degree of expectation, wherein the expectation degree acquiring means acquires the degree of expectation by using the degree of purchase possibility acquired by the possibility degree acquiring means. The product recommendation device according to claim 1. 3. The time-series relation estimating means, further comprising: a matrix generating means for generating an adjacency matrix capable of specifying a time-series order in which a user has purchased a product according to a position of a matrix element. The product recommendation apparatus according to claim 2, wherein the means specifies a related product in a time-series order relationship with the specific product from a position of a matrix element of the matrix generated by the matrix generation means. . 4. The method according to claim 1, wherein the adjacency matrix generated by the matrix generation unit further includes the user having purchased the specific product specified from the position of the matrix element by the value of the matrix element. Represents the degree of purchasing the related product specified from the position of, the likelihood degree obtaining means, after the user purchases the related product using the value of the matrix element of the adjacent matrix,
The product recommendation apparatus according to claim 2 or 3, wherein a degree of purchase possibility of purchasing another product is acquired. 5. A purchase frequency acquisition means for acquiring a purchase frequency of each product from a purchase history recorded by said recording means, wherein said expectation degree acquisition means is acquired by said purchase frequency acquisition means. The expected degree of each of the related products is acquired by using the purchase frequency obtained and the degree of purchase possibility acquired by the possibility degree acquisition means. A product recommendation device according to item 1. 6. The product information obtaining means obtains product information from the user via a network, and the recommendation means recommends product information to the user via the network. The product recommendation device according to any one of claims 1 to 5. 7. A first step of obtaining product information on a product from a user, a second step of specifying a product from the product information obtained in the first step, and a specifying in the second step A third step of estimating the related product having the specified product and the related product having a chronological order relation with respect to purchase; and a third degree of expectation that the user purchases the related product after purchasing the specific product. A fourth step of acquiring for each related product estimated in the step; and, among the related products, preferentially recommending to the user the product information of the related product having a high degree of expectation acquired in the fourth step. A product recommendation method, comprising: a fifth step of: 8. A product information obtaining function for obtaining product information on a product from a user, a specific function for specifying a product from the product information obtained by the product information obtaining function, and a specific product specified by the specific function And a product estimating function for estimating a related product having a chronological order relation with respect to purchase; and acquiring the expected degree of purchasing the related product after the user has purchased the specific product for each of the estimated related products. And a recommendation function of preferentially recommending, to the user, product information of a related product having a high degree of expectation obtained by the expectation obtaining function among the related products. Product recommendation program for.