JP2003248750A - Purchase information processing device, purchase information clustering method and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a purchase information processing device that easily and quickly extracts a cluster regulating a relation between a customer and a commodity from the purchase history information. <P>SOLUTION: A matrix forming section 2 allocates the customer and the commodity respectively to one of rows and columns on the basis of the purchase history information indicating commodity purchase history of the respective customer stored in a purchase history storing section 1 and forms a matrix table of which predetermined indicators respectively different from each other are respectively added to a matrix element of the commodity with the history of purchase of the customer and to that of the commodity without the history of purchase of the customer. A matrix rearranging section 4 rearranges the rows of the matrix table on the basis of summation of the indicators for the respective rows and rearranges the columns on the basis of summation of the indicators for the respective columns, and extracts the cluster regulated by distribution of the indicators on the matrix table. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a purchase information processing apparatus for classifying customers and classifying products, which is used for suggesting products in accordance with a customer's hobbies and tastes in a retail business. The present invention relates to a purchase information processing apparatus, a customer information clustering method, and a program for extracting a customer group according to the possibility of purchasing a product group defined in advance according to customer purchase history data.

[0002]

2. Description of the Related Art In recent years, with the development of information and communication systems, at supermarkets, convenience stores, department stores, etc., POS records of sales of individual products have been made.
(Point of Sales) It is input by the terminal device, and the sales record is stored in the computer at each store, each region, or the head office. Specifically, each product is provided with an identifier represented by a bar code, and the POS terminal device reads the identifier to specify the product. Furthermore, for example, a card called a point card is distributed to the customer side. The customer ID of this card is read by the POS terminal device when the product is purchased. Thereby, the customer is specified.

In addition, in recent years, the Web on the Internet
Commercial transactions where the site is a store (virtual store) are widespread. In such a virtual store, not only the product can be specified, but also the customer can be specified by the login ID.

On the other hand, retailers and the like send direct mails to customers in order to increase sales, or carry out campaigns for sales promotion. In general, direct mail, etc., is costly in proportion to the number of recipients, so it is necessary to recommend and propose to customers who are likely to purchase products that they are likely to purchase.

In the above-described product recommendation / suggestion, the product recommendation / suggestion can be made by extracting a set (cluster) of customers having similar purchase patterns based on the past purchase history for each customer. It can be done easily.

There are various methods for clustering analysis for extracting a cluster for each predetermined element from certain information.
For example, in "Data Mining Techniques" by Michael JA Berry and Gordon Rinoff (Kaibundou), K-me
The ans method and the agglomeration method are introduced. When these methods are applied to customer classification and product classification, it is necessary to calculate the similarity between customers for all customers.

The IPSJ journal Vol. 42
No. 8 (August 2001), Koyanagi, Kubota,
Contributed by Nakase, "Matrix Cluster
ring: A new data mining method for CRM ”
Has proposed a method called matrix clustering. Applying this approach to customer classification and product classification requires multiple iterations to find a cluster.

[0008]

Since the conventional clustering method is performed as described above, K-means is used.
In the method and the aggregation method, it is necessary to calculate the similarity between customers for all customers. Therefore, considering that the combination of customers is proportional to the square of the number of customers, there has been a problem that the amount of calculation becomes enormous as the number of customers increases.

Further, in the matrix clustering, there is a problem that the processing time becomes long because iterative processing is performed when finding one cluster.

The present invention has been made to solve the above problems, and a purchase information processing apparatus, a customer information clustering method, and a customer information clustering method capable of reducing the amount of calculation and processing time required for extracting clusters of customer information. The purpose is to get the program.

[0011]

A purchase information processing apparatus according to the present invention includes a purchase history storage unit for storing purchase history data of products for each customer, and a customer and a product in the purchase history data in rows and columns, respectively. Purchasing information generating means which is applied as one item and generates a matrix table in which predetermined matrix index values different from each other are respectively given to the matrix elements of products for which the customer has a history of purchase and the matrix elements of products for which the customer has no history of purchase. For the matrix table, the rows are sorted based on the sum of the index values for each row, and the columns are sorted based on the sum of the index values for each column, and the distribution of the index values on the matrix table is specified. And a purchase information processing means for extracting the clusters.

In the purchase information processing apparatus according to the present invention, the purchase information generation means sets the index value given to the matrix element of the product having a history of customer purchase to 1 and gives it to the matrix element of the product having no purchase history. The index value to be set is 0.

In the purchase information processing apparatus according to the present invention, the purchase information processing means rearranges the rows in the matrix table in ascending or descending order of the sum of the index values for each row, and in ascending order of the sum of the index values for each column. Alternatively, the columns are rearranged in descending order to extract clusters defined by the distribution of index values on the matrix table.

In the purchase information processing apparatus according to the present invention, the purchase information processing means multiplies the index value of each row element by the sum of the index values for each row of the matrix table, and a row weighting matrix table for each column of the matrix table. And a column weighting matrix table that is obtained by multiplying the index values of each column element by the sum of the index values, and for the original matrix table, it corresponds to the ascending or descending order of the sum of the index values for each column of the row weighting matrix table. The columns to be sorted are sorted, and the rows corresponding to the ascending or descending order of the sum of the index values for each row of the column weighting matrix table are sorted to extract the clusters defined by the distribution of the index values on the matrix table. To do.

In the purchase information processing apparatus according to the present invention, the purchase information processing means multiplies the index value of the row element by the sum of the index values for each row of the matrix table when the sum of the index values exceeds a predetermined threshold value. , A row weighting matrix table in which a row element is given a predetermined value for a threshold value or less, and a summation index value for each column of the matrix table exceeds the predetermined threshold value A column weighting matrix table that is obtained by multiplying the index value of a column element and assigning a predetermined value to the column element for those that are less than or equal to the threshold value is generated, and the original matrix table for each column of the row weighting matrix table is generated. The columns corresponding to the ascending order or the descending order of the sum of the index values are sorted, and the rows corresponding to the ascending order or the descending order of the sum of the index values for each row of the column weighting matrix table are sorted, respectively. And extracts the cluster defined by the distribution of the index values on Le.

In the purchase information processing apparatus according to the present invention, the purchase information processing means sets the predetermined value to 0 and generates the row weighting matrix table and the column weighting matrix table.

In the purchase information processing apparatus according to the present invention, each time the purchase information processing means extracts a cluster, it creates a matrix table in which the index value defining the cluster is set to 0, and the cluster extraction processing is repeated for the matrix table. It is a thing.

The purchase information clustering method according to the present invention includes a purchase history forming step of forming purchase history data of products for each customer in a memory area, and one item in a row or column for each customer and product in the purchase history data. A matrix table forming step in which a matrix table is formed in the memory area, each matrix element having a history element purchased by a customer and a matrix element of a commodity having no history of purchase are respectively given different index values. For the matrix table, the rows are sorted based on the sum of the index values for each row, and the columns are sorted based on the sum of the index values for each column, and the distribution of the index values on the matrix table is specified. And a clustering step for extracting the clusters.

The program according to the present invention applies the purchase history storage means for storing the purchase history data of the product for each customer, the customer and the product in the purchase history data are applied as one of the items in the row and the column, and the customer purchases them. Purchasing information generating means for generating a matrix table in which a predetermined index value different from each other is assigned to the matrix element of the product having a history and the matrix element of the product having no history of purchase, and the index value of each row for the matrix table As a purchase information processing means that sorts the rows based on the sum of the index values of each column and sorts the columns based on the sum of the index values of each column, and extracts the clusters defined by the distribution of the index values on the matrix table. It makes a computer work.

[0020]

BEST MODE FOR CARRYING OUT THE INVENTION An embodiment of the present invention will be described below. Embodiment 1. 1 is a diagram showing the configuration of a purchase information processing apparatus according to Embodiment 1 of the present invention. In FIG.
Reference numeral 1 denotes a purchase history recording unit (purchase history storage unit), which stores purchase history information indicating the purchase history of all customers. Reference numeral 2 denotes a matrix generation unit (purchase information generation means), which is a product (product name) purchased with each customer based on the purchase history information.
A matrix (customer-commodity matrix table) representing the relationship with is generated. A matrix storage unit (purchasing information generating means) 3 stores the matrix generated by the matrix generating unit 2. A matrix rearrangement unit (purchasing information processing means) 4 rearranges the products stored in the matrix recording unit 3 according to the sum of each row and the sum of each column. The purchase history recording unit 1 and the queue storage unit 3 form and store purchase history information in, for example, a hard disk device installed in a computer device functioning as a purchase information processing device of the present invention or another external storage device. It is embodied by the program. Further, the matrix generation unit 2 and the matrix rearrangement unit 4 are embodied by a program executed by a calculation unit such as a CPU of a computer device that functions as a purchase information processing device of the present invention.

Next, the operation will be described. First, POS
Sales data input from a terminal device (not shown) or the like is accumulated in the purchase history storage unit 1 as purchase history information (purchase history forming step). Specifically, the program functioning as the purchase history storage unit 1 creates purchase history information from the input data and records it in the memory area of the computer device. FIG. 2 is a diagram showing an example of purchase history information stored in the purchase history storage unit in FIG. As shown in the figure, the purchase history information is at least an ID that identifies a customer.
It has a (customer ID) and an ID (product ID) that identifies a product. It also includes information such as the date and time when the product was purchased, the unit price, the number of purchases, and the amount (total amount).

The matrix generation unit 2 reads the purchase history information from the purchase history storage unit 1 in a predetermined order (for example, oldest first), and the customer (customer ID) and product (product I) in the purchase history information.
A matrix (customer-product matrix table) representing the relationship between the customer (customer ID) and the product (product ID) is generated based on the combination with D) and stored in the matrix storage unit 3 (matrix table forming step). . The customer-commodity matrix table is information that allows each matrix element to be specified in association with a two-dimensional coordinate value. That is, the coordinate values on the two-dimensional coordinate plane corresponding to the customer-commodity matrix table specify the row and column items and the index value of the matrix element at which they intersect.

FIG. 3 is a diagram showing an example of a matrix generated by the matrix generator in FIG. As shown in the figure, this matrix represents products in the column direction and customers in the row direction. Here, when the index value “1” is given to each element (matrix element) in the matrix, it indicates that the customer corresponding to the row of the element purchases the product corresponding to the column. On the other hand, when the index value “0” is given to each element in the matrix, it indicates that the customer corresponding to the row of the element has never purchased the product corresponding to the column. For example, the customer B has a history of purchasing the products “a”, “c”, and “e to h”, but has no history of purchasing the products “b”, “d”, “i”, and “j”. It will be. The index values given to the matrix elements are not limited to 1 and 0, and may be different values.

Next, when the matrix generator 2 generates the above-mentioned matrix, it counts the total sum for each row and each column. For example, when the purchase history information that the customer “A” has purchased the product “a” is read from the purchase history storage unit 1, the corresponding element in the customer-product matrix (the customer is “A” and the product is “a”). The element is set to "1", 1 is added to the sum of rows (customers) "A", and 1 is added to the sum of columns (products) "a".

With respect to the matrix generated and stored in the matrix storage unit 3 as described above, the matrix rearrangement unit 4 rearranges the rows in the order of the total sum for each row and the columns in the order of the total sum for each column. Sorting is performed to generate a sorting matrix.

FIG. 4 is a diagram showing an example of the rearrangement matrix obtained by the purchase information processing apparatus according to the first embodiment of the present invention. As shown in the figure, the matrix rearrangement unit 4 rearranges the rows so that, for example, the one with the larger total sum becomes higher, and the rearrangement with the column has the larger total sum. Rearrange so that the left side is on the left side. Either of the rearrangement of rows and the rearrangement of columns may be performed first.

In the matrix thus rearranged, as shown in the figure, each element is "1" toward the upper left side.
Is higher, and the ratio of each element is “0” is higher toward the lower right side. Utilizing this, the matrix rearrangement unit 4 can select a product with a high purchase possibility and a customer with a high purchase intention by extracting the portion located on the upper left side in the above matrix as a cluster ( Clustering step). As described above, the coordinate values on the two-dimensional coordinate plane corresponding to the customer-commodity matrix table specify the row and column items and the index value of the matrix element at which they intersect. Thereby, if the coordinate values on the two-dimensional coordinate plane corresponding to the customer-product matrix table are obtained while paying attention to the index value, the coordinate value group regarding the cluster can be obtained.

Thereafter, the classification data of the purchase information is generated based on the clusters defining the products with high purchase possibility and the customers with high purchase intention, which are extracted by the matrix rearrangement unit 4, and a display device (not shown) is generated. It may be presented to the user by, for example.

As described above, according to the first embodiment, a matrix (customer-product matrix table) showing the relationship between products and customers is generated from the purchase history information, and the sum total for each row and each column is generated. The rows and columns are rearranged based on, and the clusters defined by the distribution of the index values of the matrix elements on the matrix are extracted, so cluster extraction can be performed easily and at high speed without requiring complicated calculations. You can

Embodiment 2. In the above-described first embodiment, an example in which each customer and each product are treated equally, that is, the rearrangement matrix is generated evenly, is shown, but in the second embodiment, each customer and each product are weighted. , To generate a rearrangement matrix.

The basic configuration of the purchase information processing apparatus according to the second embodiment is the same as that of the first embodiment,
The function of the matrix rearrangement unit 4 is different from the function described in the first embodiment. Further, it is assumed that the matrix storage unit 3 stores the matrix (customer-product matrix table) described in the first embodiment.

Next, the operation will be described. Here, a matrix rearrangement process different from the function described in the first embodiment will be described, and the other operations are the same as those in the first embodiment, and thus the description thereof will be omitted. First, the matrix rearrangement unit 4
Generates a weighting matrix for rows and columns for the matrix (customer-commodity matrix table: see FIG. 3) stored in the matrix storage unit 3.

FIG. 5 is an explanatory view for explaining a rearrangement matrix obtained by the purchase information processing apparatus according to the second embodiment of the present invention, where (a) shows a row weighting matrix and (b).
Shows a column weighting matrix, and (c) shows the matrix after rearrangement. As shown in FIG. 5A, the matrix rearrangement unit 4 generates a row weighting matrix in which each row element is multiplied by the sum of the index values for each row of the matrix stored in the matrix storage unit 3. Similarly, as shown in FIG. 5B, a column weighting matrix in which each column element is multiplied by the sum of the index values for each column is generated.

Next, the matrix rearrangement unit 4 refers to the row weighting matrix and the column weighting matrix and, in the matrix table of the customer-commodity, responds in the order of the sum of the index values of each column in the row weighting matrix. In addition to rearranging the columns, the corresponding rows are rearranged in the order of the sum of the index values of each row in the column weighting matrix to generate a rearranged matrix as shown in FIG. 5C. Either of the rearrangement of rows and the rearrangement of columns may be performed first.

By the way, in the customer-commodity matrix table stored in the matrix storage unit 3, the total sum for each row or each column indicates the importance for each commodity or each customer. In consideration of this, each element is multiplied by the sum of these rows or the sum of columns as a weighting coefficient. By sorting in this way, even customers who have purchased the same number of products can purchase, for example, more important products (i.e., they can be purchased by a large number of customers by being located on the left side of the sorted queue). Products)
The customer who has purchased is positioned on the upper side in the figure. As a result, the ratio of the index value of each element being “1” becomes higher toward the upper left side in the figure. Clusters are extracted based on this distribution of index values.

As described above, according to the second embodiment, a matrix (customer-product matrix table) showing the relationship between the product and the customer is generated from the purchase history information, and a row is created for each element of this matrix. The weighted sum is calculated for each row and each column, and the weighted sum is obtained for each row and each column. The rows and columns of the customer-product matrix are rearranged according to these weighted sums, and matrix elements on the matrix are arranged. Since the clusters defined by the distribution of the index values are extracted, it is possible to weight the products and customers and to easily and quickly extract the clusters.

Embodiment 3. In the above-described second embodiment, an example is shown in which the sum of each row is used as a weighting coefficient for customers (customer weighting coefficient) and the sum of each column is used as a weighting coefficient for products (product weighting coefficient). Is to set a threshold for the customer weight coefficient and the product weight coefficient.

The basic structure of the purchase information processing apparatus according to the second embodiment is the same as that of the first embodiment,
The function of the matrix rearrangement unit 4 is different from the function described in the first and second embodiments. Further, the matrix storage unit 3 stores the matrix (customer-product matrix table) described in the first embodiment.
Is stored.

Next, the operation will be described. Here, a matrix rearrangement process different from the functions described in the first and second embodiments will be described, and other operations will be described in the first embodiment.
The description is omitted because it is similar to the above. First, the matrix rearrangement unit 4 determines whether the sum of index values of rows or columns corresponding to each element of the matrix exceeds a preset threshold value with respect to the customer-product matrix stored in the matrix storage unit 3. Multiplies the sum total by the index value of the row or column element, and when the sum total of the index values of the respective elements is less than or equal to the threshold value, sets the element to 0 and generates a weighting matrix.

FIG. 6 is an explanatory view for explaining a rearrangement matrix obtained by the purchase information processing apparatus according to the third embodiment of the present invention, where (a) shows a row weighting matrix and (b).
Shows a column weighting matrix, and (c) shows the matrix after rearrangement. As shown in FIG. 6A, the matrix rearrangement unit 4 sets a threshold value (shown in the figure) in which the sum of the index values of the rows corresponding to the elements of the matrix is preset for the matrix stored in the matrix storage unit 3. When it exceeds 4, the element is multiplied by the total sum of the row, and when the total sum of the index values of the row corresponding to each element of the matrix is less than or equal to the threshold value, the element is set to 0 and the row weighting is performed. Generate a matrix. Similarly, as shown in FIG. 6B, when the sum of the index values of the columns corresponding to the respective elements of the matrix exceeds a preset threshold value (2 in the illustrated example), An element is multiplied by the total sum of columns, and when the total sum of the index values of the column corresponding to each element of the matrix is equal to or less than the threshold value, the element is set to 0 to generate a column weighting matrix.

Next, the matrix rearrangement unit 4 refers to the row weighting matrix and the column weighting matrix and, in the matrix table of the customer-commodity, corresponds to the sum of the index values of each column in the row weighting matrix in this order. The columns to be sorted are rearranged, and the corresponding rows are rearranged in the order of the sum of the index values of each row in the column weighting matrix to generate a rearranged matrix as shown in FIG. 6C. Either of the rearrangement of rows and the rearrangement of columns may be performed first.

Thus, in the customer-commodity matrix, thresholds are set for the sum of rows and the sum of columns, and weighting is performed depending on whether the sum of rows and the sum of columns exceed their respective thresholds. Since it decides whether to set it to zero or zero, avoid placing the customer who purchased many unimportant products (that is, products located on the right side and purchased by only a few customers) in the upper part of the figure. You can As a result, the ratio of the index value of each element being “1” becomes higher toward the upper left side in the figure. Clusters are extracted based on this distribution of index values.

As described above, according to the third embodiment, with respect to the customer-commodity matrix, thresholds are set for the sum of rows and the sum of columns, respectively, and whether the sum of rows and the sum of columns exceed the threshold. Depending on whether or not each element is weighted or set to zero, it is possible not only to easily perform cluster extraction at high speed, but also to perform cluster extraction with high accuracy.

Fourth Embodiment In the above-described embodiment, an example in which an area in which the ratio of the elements of the customer-product matrix is “1” is high, that is, a cluster is extracted in the upper left part of the matrix, is shown. All the elements inside are set to 0, and the process is repeated again in the same manner to sequentially extract clusters.

Next, the operation will be described. First, the matrix rearrangement unit 4 extracts clusters from the rearranged matrix obtained by the operation described in the above embodiment. Then, with all the elements in this cluster set to 0, the sorting is executed again by the operation described in the above embodiment.

FIG. 7 is an explanatory view for explaining a rearrangement matrix obtained by the purchase information processing apparatus according to the fourth embodiment of the present invention, where (a) shows a row weighting matrix and (b).
Shows a column weighting matrix, and (c) shows the matrix after rearrangement. In the illustrated example, the matrix rearrangement unit 4 extracts 4 rows and 5 columns located on the upper left side as a first cluster from the rearranged matrix obtained by the operation described in the second embodiment. Then, with all the elements in the first cluster set to 0, the sorting is executed again by the method described in the second embodiment. That is, as shown in FIG. 7A, a matrix obtained by multiplying the total sum of rows corresponding to each element of the rearranged matrix in which all the elements in the first cluster are 0 is generated as a row weighting matrix. At this time, the sum is calculated for each column of the row weighting matrix.

Similarly, as shown in FIG. 7B, a column weighting matrix is obtained by multiplying the sum of the columns corresponding to each element of the weighting rearrangement matrix in which all the elements in the first cluster are zero. Generate as. Also at this time, the sum is obtained for each row of the column weighting matrix.

Next, referring to the row and column weighting matrix, the rows are rearranged in the order of the weighted sum of the customer-product matrix, the columns are rearranged in the order of the column weighted sum, and the weighted rearrangement matrix is again displayed. Is generated (Fig. 7
(See (c)). In FIG. 7C, rearrangement is performed by regarding “1” shown in italics as zero.

As described above, according to the fourth embodiment, after the first cluster is extracted, all the elements in the first cluster are set to zero, and the rearrangement is executed again. Two clusters can be easily extracted. In this way, after extracting the (N-1) th cluster, the (N-1) th cluster is extracted.
It is possible to extract the Nth cluster by executing the rearrangement again by setting all the elements in the cluster of -1) to zero.

In the above-described embodiment, an example has been described in which the cluster is extracted according to the purchase history information obtained by the customer purchasing the product. However, for example, when the customer browses a Web page or on the Internet. With respect to customer history information obtained when browsing and downloading contents such as moving images or music, cluster extraction can be similarly performed. That is, the product of the present invention is defined by a concept including information such as a Web page and the above content.

[0051]

As described above, according to the present invention, the purchase history data of the product for each customer is stored, and the customer and the product in the purchase history data are applied as one item of the row and column, respectively. A matrix table is created by assigning different index values to the matrix elements of products that have a history of customer purchases and the matrix elements of products that do not have a history of purchases. The rows are sorted based on the sum of the values, and the columns are sorted based on the sum of the index values for each column to extract the clusters defined by the distribution of the index values on the matrix table. There is an effect that can be performed easily and at high speed.

According to the present invention, the index value given to the matrix element of the product having a history of purchase by the customer is set to 1, and the index value given to the matrix element of the product having no purchase history is set to 0. There is an effect that the amount of calculation at the time of extraction is reduced and cluster extraction can be performed at high speed.

According to the present invention, in the matrix table, the rows are sorted in the ascending or descending order of the sum of the index values for each row, and the columns are sorted in the ascending or descending order of the sum of the index values for each column. Since the clusters defined by the distribution of the index values on the matrix table are extracted, there is an effect that the calculation at the time of cluster extraction becomes simple.

According to the present invention, the row weighting matrix table formed by multiplying the sum of the index values for each row of the matrix table by the index value of each row element, and the sum of the index values for each column of the matrix table of each column element And a column weighting matrix table that is obtained by multiplying the index values, and rearranges the corresponding columns in the ascending or descending order of the sum of the index values for each column of the row weighting matrix table with respect to the original matrix table, and the column weighting matrix The corresponding rows are sorted in ascending or descending order of the sum of the index values for each row in the table, and the clusters defined by the distribution of the index values on the matrix table are extracted, so weighting is applied to products and customers. Moreover, there is an effect that cluster extraction can be performed easily and at high speed.

According to the present invention, when the sum of the index values for each row of the matrix table exceeds a predetermined threshold value, the sum of the index values of the row elements is multiplied by the sum. Row weighting matrix table in which a predetermined value is added to the matrix table, and the sum of the index values for each column of the matrix table exceeds a predetermined threshold, the sum is multiplied by the index value of the column element For the following, a column weighting matrix table in which column elements are given predetermined values is generated, and the original matrix table is sorted in ascending or descending order of the sum of the index values for each column of the row weighting matrix table. In addition to rearranging the corresponding columns, rearranging the corresponding rows in ascending or descending order of the sum of the index values for each row of the column weighting matrix table, the columns specified by the distribution of the index values in the matrix table are sorted. Since extracting the static, it is possible not only to perform the cluster extraction easily at high speed, there is an effect that it is possible to accurately perform cluster extraction according to the threshold.

According to the present invention, the row weighting matrix table and the column weighting matrix table are generated by setting the predetermined value to 0, so that there is an effect that cluster extraction can be performed easily and accurately.

According to the present invention, each time a cluster is extracted, a matrix table in which the index value defining this is set to 0 is generated, and the cluster extraction process is repeated for the matrix table. The effect is that it can be done.

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration of a purchase information processing apparatus according to a first embodiment of the present invention.

FIG. 2 is a diagram showing an example of purchase history information stored in a purchase history storage unit in FIG.

FIG. 3 is a diagram showing an example of a matrix generated by a matrix generator in FIG.

FIG. 4 is a diagram showing an example of a rearrangement matrix obtained by the purchase information processing apparatus according to the first embodiment of the present invention.

FIG. 5 is an explanatory diagram illustrating a rearrangement matrix obtained by the purchase information processing apparatus according to the second embodiment of the present invention.

FIG. 6 is an explanatory diagram illustrating a rearrangement matrix obtained by the purchase information processing apparatus according to the third embodiment of the present invention.

FIG. 7 is an explanatory diagram illustrating a sorting matrix obtained by a purchase information processing apparatus according to Embodiment 4 of the present invention.

[Explanation of symbols]

1 purchase history storage unit (purchase history storage unit), 2 matrix generation unit (purchase information generation unit), 3 matrix storage unit (purchase information generation unit), 4 matrix rearrangement unit (purchase information processing unit).

Claims (9)

[Claims] 1. A purchase history storage means for storing purchase history data of a product for each customer, and a history of purchase by a customer by applying the customer and the product in the purchase history data as one of items in rows and columns, respectively. Purchasing information generating means for generating a matrix table in which predetermined matrix index values that are different from each other are assigned to the matrix elements of products and the matrix elements of products for which there is no history of purchase; A purchase information processing unit that sorts the rows based on the total sum, sorts the columns based on the total sum of the index values for each column, and extracts the clusters defined by the distribution of the index values on the matrix table. Purchasing information processing device equipped. 2. The purchase information generating means sets an index value of 1 to a matrix element of a product that a customer has a history of purchasing, and sets an index value of 0 to a matrix element of a product that has no history of purchase by a customer.
The purchase information processing apparatus according to claim 1, wherein: 3. The purchase information processing means sorts the rows of the matrix table in ascending or descending order of the sum of index values for each row, and sorts the columns in ascending or descending order of the sum of index values for each column. The customer information processing apparatus according to claim 1 or 2, wherein a cluster defined by a distribution of index values on the matrix table is extracted. 4. The purchase information processing means calculates a sum of index values for each row of the matrix table by multiplying the index value of each row element by a row weighting matrix table, and a sum of index values for each column of the matrix table. A column weighting matrix table generated by multiplying the index values of the column elements is generated, and the columns corresponding to the sum of the index values of each row of the row weighting matrix table are rearranged in the original matrix table, respectively, and the corresponding columns are rearranged. Sorting the corresponding rows in ascending or descending order of the sum of the index values for each row of the column weighting matrix table,
The cluster defined by the distribution of index values on the matrix table is extracted.
Customer information processing device described. 5. The purchase information processing means multiplies the sum of the index values of row elements by the sum of the index values for each row of the matrix table when the sum of the index values exceeds a predetermined threshold value. A row weighting matrix table in which a row element is given a predetermined value, and when the sum of index values for each column of the matrix table exceeds a predetermined threshold value, the index value of the column element is multiplied by the sum. , A column weighting matrix table in which column elements are given the above-specified values for those below the threshold, and the sum of the index values for each column of the row weighting matrix table is generated for the original matrix table. The columns corresponding to each of the ascending order and the descending order are rearranged, and the rows corresponding to the sum of the index values for each row of the column weighting matrix table are rearranged in the ascending order or the descending order, respectively, and the corresponding index value on the matrix table is divided. The customer information processing apparatus according to claim 1 or 2, wherein clusters defined by cloth are extracted. 6. The customer information processing apparatus according to claim 5, wherein the purchase information processing means sets a predetermined value to 0 and generates a row weighting matrix table and a column weighting matrix table. 7. The purchase information processing means generates a matrix table in which an index value defining the cluster is set to 0 every time a cluster is extracted, and the cluster extraction processing is repeated for the matrix table. 7. The customer information processing device according to claim 6. 8. A purchase history forming step of forming purchase history data of a product for each customer in a memory area, and applying the customer and the product in the purchase history data as one item of row and column, respectively, and purchasing by the customer A matrix table forming step of forming in the memory area a matrix table in which a predetermined index value different from each other is assigned to the matrix element of the product having a history and the matrix element of the product having no history of purchase, and the matrix table, The rows are sorted based on the sum of the index values for each row, and the columns are sorted according to the sum of the index values for each column to extract the clusters defined by the distribution of the index values on the matrix table. A purchasing information clustering method comprising a clustering step. 9. A purchase history storage means for storing purchase history data of a product for each customer, a product having a history of purchase by applying the customer and the product in the purchase history data as one of items in a row and a column, respectively. Purchase information generating means for generating a matrix table in which predetermined matrix index values that are different from each other are assigned to the matrix elements of and the matrix elements of products for which there is no history of purchase. The computer functions as a purchase information processing unit that sorts the rows based on the sum of the index values for each column, and extracts the clusters defined by the distribution of the index values on the matrix table. Program to let.