WO2024195063A1 - Analysis device - Google Patents

Abstract

This analysis device includes a similarity calculation unit that accepts flow line information and purchasing information for each customer, and calculates the similarity between flow lines for each combination of flow lines on the basis of a flow line shape obtained from the flow line information and a purchasing trend obtained from the purchasing information.

Description

Analytical Equipment

The disclosed technology relates to analyzing customer behavior in retail stores and supporting policy planning.

As sensor devices become cheaper and recognition technology becomes more accurate, there has been a growing trend in recent years to visualize customer movements within stores and increase sales.

Changing customer movement patterns can change customer purchasing behavior, and it is known that the length of movement patterns in particular correlates with the number of purchases (for example, non-patent literature 1).

In a related-art initiative, efforts have been made to analyze information on customers' time spent in front of shelves and move the sales area for specific products to guide them to the back of the store and increase the flow of customers (Non-Patent Document 2).

Shimizu, Satoshi, A new perspective on traffic flow research, Marketing Journal 40 (2), 7-17, 2020-09-29, Japan Marketing Association. Nikkei Crosstech: Increase the number of customers guided to the back of the store by 20%, analyzing in-store videos with Edge (2017), URL: https://xtech.nikkei.com/it/atcl/column/17/090600369/090600004/

However, conventional technology does not take into account differences in purchasing characteristics of each customer or the characteristics of each customer's movement path, so there is a possibility that customers will be guided along a movement path that is difficult to actually follow.

For example, Figures 17A to 17C show examples of customer movement patterns for each unit price. Figure 11A shows the movement pattern of a low-price customer who purchases only one item, "coffee," Figure 17B shows the movement pattern of a medium-price customer who purchases two items, "coffee and sweets," and Figure 17C shows the movement pattern of a high-price customer who purchases three items, "lunch box, tea, and sweets."

If we only consider the unit price, the recommended pattern will be the movement of a high-priced customer who buys three items and travels around the entire store compared to a low-priced customer who buys only one coffee item.

However, the purchasing trends (single coffee) and movement patterns (short movement pattern on the left side of the store) of low-price customers shown in Figure 17A are significantly different from the purchasing trends and movement patterns of high-price customers shown in Figure 17C. Therefore, it is expected that the number of customers who will move from the former movement pattern to the latter movement pattern will be small, or that it will be very costly to induce them. Instead, it is thought that a more cost-effective and realistic proposal would be to recommend medium-price customers who have similar purchasing trends and movement patterns.

The disclosed technology has been developed in consideration of the above points, and aims to provide an analysis device that enables analysis that takes into account the characteristics of customer purchases and movement patterns.

A first aspect of the present disclosure is an analysis device that includes a similarity calculation unit that receives purchase information and movement line information for each customer, and calculates the similarity between movement lines for each combination of movement lines based on the movement line shape obtained from the movement line information and the purchasing tendency obtained from the purchase information.

The disclosed technology makes it possible to perform analysis that takes into account the characteristics of customer purchases and movement patterns.

FIG. 1 is a block diagram showing the hardware configuration of the analysis device. FIG. 2 is a block diagram showing the configuration of the analysis device of the first embodiment. FIG. 3A is a diagram showing an example of purchase information for each customer. FIG. 3B is a diagram illustrating an example of flow line information for each customer. FIG. 4A is a diagram showing a sequence of customers passing through an area divided into an arbitrary shape. FIG. 4B is a diagram showing a user's trajectory as a series of areas. FIG. 5 is a diagram showing an example of a graph representing the space inside a store. FIG. 6 is a diagram showing an example of a two-dimensional map representing purchase information associated with flow lines and distances between the flow lines. FIG. 7 is a flowchart showing the flow of the analysis process by the analysis device 100 of the first embodiment. FIG. 8 is a block diagram showing the configuration of the analysis device of the second embodiment. FIG. 9 is a flowchart showing the flow of an analysis process performed by the analysis device of the second embodiment. FIG. 10 is a block diagram showing the configuration of an analysis device according to the third embodiment. FIG. 11 is a flowchart showing the flow of an analysis process performed by the analysis device of the third embodiment. FIG. 12 is a block diagram showing the configuration of the analysis device of the fourth embodiment. FIG. 13 is a diagram showing an example of an axis representing a flow line shape cluster and an axis representing a purchase cluster. FIG. 14 is a flowchart showing the flow of an analysis process performed by the analysis device of the fourth embodiment. FIG. 15 is a diagram showing the mutual information divided between grouping pattern A and grouping pattern B. FIG. 16 is a diagram comparing the histograms of grouping pattern A and grouping pattern B. FIG. 13 is a diagram showing an example of the flow of low-price customers. FIG. 13 is a diagram showing an example of the flow of customers with medium unit prices. FIG. 13 is a diagram showing an example of the flow of high-priced customers.

Below, an example of an embodiment of the disclosed technology will be described with reference to the drawings. Note that the same reference symbols are used for identical or equivalent components and parts in each drawing. Also, the dimensional ratios in the drawings have been exaggerated for the convenience of explanation and may differ from the actual ratios. Note that below, explanations common to each embodiment will be referred to as this embodiment.

The technology disclosed herein finds the degree of similarity between traffic lines based on differences in the traffic line shape, purchasing trends, etc., and after compressing the dimensions, links it to the purchase price and the number of orders linked to each traffic line, etc., and displays it to users such as store managers. This allows users to consider measures while taking into account the degree of similarity between traffic lines.

By presenting this information, it is possible to grasp, for example, which traffic flow will contribute most to increasing sales. The method of this embodiment can also solve the problem that it is not possible to grasp which measures are effective based on sales alone, and it is not possible to grasp which traffic flow should be guided based on the similarity between measures alone.

The configuration of this embodiment is explained below.

FIG. 1 is a block diagram showing the hardware configuration of the analysis device 100.

As shown in FIG. 1, the analysis device 100 has a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, storage 14, an input interface (I/F) 15, a display interface (I/F) 16, and a communication interface (I/F) 17. Each component is connected to each other so as to be able to communicate with each other via a bus 19.

The CPU 11 is a central processing unit that executes various programs and controls each part. That is, the CPU 11 reads the programs from the ROM 12 or storage 14, and executes the programs using the RAM 13 as a working area. The CPU 11 controls each of the above components and performs various calculation processes according to the programs stored in the ROM 12 or storage 14. In this embodiment, an analysis program is stored in the ROM 12 or storage 14.

ROM 12 stores various programs and data. RAM 13 temporarily stores programs or data as a working area. Storage 14 is composed of a storage device such as an HDD (Hard Disk Drive) or SSD (Solid State Drive), and stores various programs including an operating system, and various data.

The input interface 15 includes a pointing device such as a mouse and a keyboard, and is used to perform various input operations.

The display interface 16 is, for example, a liquid crystal display, and displays various information. The display interface 16 may also function as the input interface 15 by adopting a touch panel system.

The communication interface 17 is an interface for communicating with other devices such as terminals. For this communication, for example, a wired communication standard such as Ethernet (registered trademark) or FDDI, or a wireless communication standard such as 4G, 5G, or Wi-Fi (registered trademark) is used.

[First embodiment]
Next, each functional configuration of the analysis device 100 of the first embodiment will be described. Fig. 2 is a block diagram showing the configuration of the analysis device of the first embodiment. Each functional configuration is realized by the CPU 11 reading out an analysis program stored in the ROM 12 or the storage 14, expanding it in the RAM 13, and executing it.

As shown in FIG. 2, the analysis device 100 includes an information database (DB) 102, a similarity calculation unit 110, and a display unit 112.

Information DB 102 holds two types of information: purchasing information and traffic flow information for each customer. Purchasing information for each customer consists of a customer ID, order ID, store entry and exit times, the customer's purchased items, the number of items, and the order of purchases. Traffic flow information for each customer consists of a customer ID, order ID, store entry and exit times, traffic flow ID, and traffic flow route information.

The purchasing information and the traffic flow information for each customer may be stored as a single table or as two tables, with the customer ID and order ID as common keys. FIG. 3A is a diagram showing an example of purchasing information for each customer. In the example shown in FIG. 3A, the customer ID, order ID, store entry and exit times, purchased items, purchase order, and purchase unit price are illustrated. FIG. 3B is a diagram showing an example of traffic flow information for each customer. In the example shown in FIG. 3B, the customer ID, order ID, traffic flow ID, traffic flow path information, and shaped traffic flow information are illustrated. The traffic flow path information included in the traffic flow information represents the trajectory taken by each customer within the store. Note that the traffic flow path information between traffic lines or the shaped traffic flow information is an example of information representing the traffic flow shape of the present disclosure, and at least one of the purchased items, purchase order, and purchase unit price is an example of the purchasing tendency of the present disclosure. In the processing of the similarity calculation unit 110 described below, in the first embodiment, a pair of flow line route information is used as information representing the flow line shape, and in the second embodiment, a pair of shaped flow line information is used as information representing the flow line shape. Also, Figures 3A and 3B are just examples, and some information is omitted.

The traffic line route information may be time-series coordinate data showing the trajectory of the customer, or may be information on the series of areas that the customer passed through, which are divided into areas of any shape, as shown in FIG. 4A. In FIG. 4B, the trajectory of the user shown in FIG. 4A is represented as a series of areas. Details of FIG. 4B will be described later in the second embodiment.

The similarity calculation unit 110 receives purchase information and movement line information for each customer from the information DB 102, and calculates the similarity between movement lines for each combination of movement lines based on the movement line shape obtained from the movement line information and the purchasing tendency obtained from the purchase information. The closer the distance between each movement line, the higher the similarity. There are several methods for calculating the similarity between two movement lines, and the following methods (a) to (d) are shown as examples.
Note that (a) and (b) are examples of cases where the similarity between traffic lines is calculated from the traffic line shapes, and (c) is an example of cases where the similarity between traffic lines is calculated from purchasing tendencies. As will be explained below, purchasing tendencies are also calculated as distances.

The output of the similarity calculation unit 110 may be a single value of the distance between two traffic lines calculated using one of the above methods, or multiple values that combine similarities calculated from traffic line shapes using multiple methods and similarities calculated from purchasing trends.

(a): Calculation of distance by DTW In this method, a pair of two flow line route information is input, and it can be performed by, for example, dynamic time warping (DTW). In DTW, the distance between each point of two time series data is measured by exhaustive calculation, and the minimum value of the sum of the corresponding distances between each point is set as the distance (=similarity) between the two time series data. This method is characterized by measuring the distance even in time series of different lengths, and can calculate the distance even when the path length is different. For example, when comparing a short flow line and a long flow line, the path length is different. When calculating the distance between each point, Euclidean distance may be used, or an in-store distance considering the path in the store may be used. When calculating the in-store distance, the space in the store is divided (for example, divided into a grid) as a preprocessing, and the grids at the places where people can pass in the store are connected as nodes of a graph, and adjacent grids are connected by links, and the in-store space is represented in a graph. FIG. 5 is a diagram showing an example of an in-store space represented in a graph. The dotted lines represent shelves, and the left side of Fig. 5 is a graph showing the in-store space, while the right side of Fig. 5 is a graph showing coordinates A and B. On the right side, the routes selected for coordinates A and B are indicated with arrows. When calculating the in-store distance between coordinates A and B, the nodes closest to coordinates A and B are selected, respectively, and the route length of the shortest route connecting these two nodes is found, and this route length is regarded as the in-store distance. By using such in-store distances, it is possible to calculate distances that are appropriate to the actual conditions inside the store, including constraints such as shelves and other obstacles.

(b): Distance calculation by edit distance This method is used when each piece of flow line route information is made up of area series information. This method calculates the edit distance by inputting a pair of two pieces of flow line route information. For example, two pieces of flow line route information are input, and the edit distance between the two pieces of area series information is measured, and the edit distance of the measurement result may be used as the similarity. For example, when the areas that transition along a route are represented by numbers, the edit distance between the flow line route "0 → 1 → 0" and the flow line route "0 → 1 → 2 → 1 → 0" is equal to the cost of inserting area 2 into the route. The cost may be calculated using the Euclidean distance or the inverse of the transition probability between areas.

(c): Calculation of distance based on purchasing tendency This method measures the similarity between the flow lines according to the difference in purchasing tendency based on the purchasing information in each flow line. The purchased items linked to each flow line are expressed as vectors, and the distance between the vectors can be calculated. In addition, the flow line and the purchased items can be linked by a method such as taking the average of the vectors of multiple purchases along a certain flow line. For example, the purchased item of flow line A is coffee, and the purchased items of flow line B are tea and lunch boxes. In this case, the purchase vector of flow line A is [1,0,0], and the purchase vector of flow line B is [0,1,1]. Each component of the vector represents the purchase of coffee, tea, and lunch boxes. The distance between flow line A and flow line B can be measured as 3, for example, when calculated using the Manhattan distance.

(d): Adding up the distances of (a) to (c) The similarity calculation unit 110 adds up the similarities calculated in (a) to (c) and defines the sum as the similarity. At this time, each value may be weighted or scaled before being added up.

The display unit 112 arranges and displays each of the movement lines associated with purchasing information for each movement line in a two-dimensional space based on the similarity between the movement lines, so that the similarity between the movement lines is represented as a two-dimensional distance based on the similarity between the movement lines. For example, each movement line is arranged on a two-dimensional map based on the similarity between the movement lines. The display unit 112 combines the similarity between each movement line obtained from the similarity calculation unit 110 and the purchasing information for each movement line and displays it as a two-dimensional map. To generate the two-dimensional map, a method such as SNE (Stochastic Neighbor Embedding) is used to compress the relationship between each movement line into two dimensions while retaining information on the distance between the lines.

6 is a diagram showing an example of a two-dimensional map that represents purchase information associated with the flow lines and the distance between the flow lines. As shown in FIG. 5, the similarity of the flow lines is represented by the distance in two dimensions, and the purchase information of each flow line, such as the purchase price and the number of orders, is represented by the size and shade of the circle. In the example of FIG. 6, the size of the circle for each flow line is represented by the number of orders (N). The purchase price is also represented by the shade of the color of the circle. The darker the circle, the higher the purchase price. In this way, the display unit 112 arranges each of the flow lines in two dimensions with a size according to a predetermined numerical value included in the purchase information. Flow line A has a purchase price of 120 yen and a number of orders (N) of 150. Flow line B has a purchase price of 300 yen and a number of orders (N) of 200. Flow line C has a purchase price of 200 yen and a number of orders (N) of 250. Flow line D has a purchase price of 500 yen and a number of orders (N) of 200. Flow line E has a purchase price of 100 yen and an order number (N) of 100. The higher the similarity between the flow lines, the closer the flow lines are located, and the lower the similarity between the flow lines, the farther the flow lines are located. In addition, the display unit 112 presents each of the flow lines arranged in two dimensions in association with the purchase price of the purchasing information for the flow line and the order number (N) for the flow line, by using a speech bubble or other representation of the flow line.

The above-mentioned two-dimensional map representation displayed by the display unit 112 allows users such as store managers to easily find the target traffic flow (e.g., low unit purchase price) and the destination traffic flow (e.g., high unit purchase price), and enables them to consider measures while checking the similarity.

Next, the operation of the analysis device 100 of the first embodiment will be described. FIG. 7 is a flowchart showing the flow of analysis processing by the analysis device 100 of the first embodiment. The analysis processing is performed by the CPU 11 reading out an analysis program from the ROM 12 or storage 14, expanding it into the RAM 13, and executing it. The analysis device 100 performs the following processing using the purchasing information and movement line information for each customer that have been acquired in advance from the information DB 102 as input.

In step S100, the CPU 11, as the similarity calculation unit 110, receives purchasing information and movement line information for each customer from the information DB 102.

In step S102, the CPU 11, as the similarity calculation unit 110, calculates the similarity between the movement lines for each combination of movement lines based on the movement line shape obtained from the movement line information and the purchasing tendency obtained from the purchasing information.

In step S104, the CPU 11 causes the display unit 112 to display each of the flow lines associated with purchasing information for each flow line in a two-dimensional layout based on the similarity between the flow lines, so that the similarity between the flow lines is represented as a two-dimensional distance.

As described above, the analysis device 100 of the first embodiment enables analysis that takes into account the characteristics of customer purchases and movement lines. Furthermore, by displaying a two-dimensional map that is based on an analysis that takes into account the characteristics of customer purchases and movement lines, the user can consider measures that take into account the characteristics of customer purchases and movement lines.

[Second embodiment]
8, the analysis device 100 of the second embodiment further includes a movement line shaping unit 210 as a preprocessing unit of the similarity calculation unit 110. The analysis device 100 of the second embodiment includes an information DB 102, the movement line shaping unit 210, the similarity calculation unit 110, and a display unit 112.

The flow line shaping unit 210 accepts the flow line route information of the flow line information in the information DB 102, and adds the shaped information to the flow line information as shaped flow line information. The flow line shaping unit 210 divides the series in the flow line route information into several chunks and converts them into shorter route information. For example, in FIG. 4B, the series information of the area through which the customer passed is converted into location information in the store (a label indicating the area name). In the example shown in FIG. 4B, area 0 corresponds to the entrance, areas 1, 2, and 3 correspond to the left aisle, and area 4 corresponds to the upper left. In this case, the shaped flow line information is "entrance → left aisle → upper left → left aisle → entrance", and it can be seen that it has been converted into shorter route information compared to when the series is expressed by numbers. This improves the readability of the flow line route information and shortens the length of the series information, making it possible to speed up the similarity calculation process. In the second embodiment, the similarity calculation unit 110 described in the first embodiment uses reshaped flow line information instead of flow line route information as information representing the flow line shape in the process of calculating the similarity. As described above, the flow line reshaping unit 210 divides the series of flow line route information in the flow line information into area units and converts it into reshaped route information on the area. Note that in the reshaped route information, a pattern of a continuous/overlapping sequence of the same area may be converted into a single pattern by deleting the overlaps (for example, converting area ABABAB to area AB). This leads to further speeding up the calculation process.

FIG. 9 is a flowchart showing the flow of analysis processing by the analysis device 100 of the second embodiment. In the second embodiment, steps S200 and S202 are executed before step S100.

In step S200, the CPU 11, as the movement line shaping unit 210, divides the sequence of movement line route information in the movement line information into area units and converts it into shaped route information on the area. After processing step S200, the process proceeds to step S202.

In step S202, the CPU 11, functioning as the traffic line shaping unit 210, adds the converted shaped route information to the traffic line information and stores it in the information DB 102.

Step S100 and subsequent steps are the same as in the first embodiment, but differ from the first embodiment in that shaped path information is used to calculate the similarity.

As described above, the analysis device 100 of the second embodiment makes it possible to speed up the similarity calculation process in an analysis that takes into account the characteristics of customer purchases and traffic patterns.

[Third embodiment]
As shown in FIG. 10, the analysis device 100 of the third embodiment further includes an extraction unit 310 as a pre-processing unit of the similarity calculation unit 110. The analysis device 100 of the third embodiment is configured to include an information DB 102, a flow line shaping unit 210, an extraction unit 310, a similarity calculation unit 110, and a display unit 112. The extraction unit 310 is assumed to extract flow line route information or shaped route information, but the following description will be given taking the case of extracting flow line route information as an example. In addition, the similarity calculation unit 110 may perform similarity calculation using the narrowed down representative flow line route information and shaped route information, respectively. The flow line route information and shaped route information are examples of information related to the flow line route of the present disclosure.

The extraction unit 310 extracts representative movement line route information from the movement line route information included in the movement line information, and outputs the representative movement line route information to the similarity calculation unit 110. Alternatively, the extraction unit 310 converts multiple pieces of movement line route information in the movement line information into representative movement line route information, and outputs the movement line information to the similarity calculation unit 110.

One possible method for extracting the main traffic lines is to determine which routes appear frequently as the main traffic lines. Another possible method is to use a clustering method such as K-medoids to divide a large amount of traffic line route information and extract representative traffic lines (main traffic lines) for each divided group.

FIG. 11 is a flowchart showing the flow of analysis processing by the analysis device 100 of the third embodiment. In the third embodiment, step S300 is executed after step S100.

In step S300, the CPU 11, functioning as the extraction unit 310, extracts representative movement line route information from the movement line route information included in the movement line information, and outputs the representative movement line route information to the similarity calculation unit 110, which processes the next step 102.

As described above, the analysis device 100 of the third embodiment makes it possible to speed up the similarity calculation process by narrowing down the routes in an analysis that takes into account the characteristics of customer purchases and traffic lines.

[Fourth embodiment]
12 , the analysis device 100 of the fourth embodiment further includes a clustering unit 410 and a cluster display unit 412 instead of the display unit 112. In the fourth embodiment, the processing of the clustering unit 410 is disposed between the similarity calculation unit 110 and the cluster display unit 412.

The clustering unit 410 classifies multiple similar movement lines according to a predetermined clustering method based on the similarity between the movement lines received from the similarity calculation unit 110, and clusters the movement lines. Note that the clustering unit 410 receives both the similarity based on the movement line shapes related to (a) and (b) above, and the similarity based on the purchasing tendency related to (c). Unlike the above-mentioned embodiment, by grouping multiple similar movement lines, the number of objects that the user/administrator needs to check can be reduced, reducing the burden.

The clustering method of the clustering unit 410 will be explained. The clustering unit 410 performs clustering based on the similarity of the movement line shape and the purchasing tendency. For example, the clustering method uses non-hierarchical clustering such as K-medoids or hierarchical clustering such as Ward's method. As a result, each movement line belongs to a cluster based on the movement line shape (hereinafter, movement line shape clusters 1 to M) and a cluster based on the purchasing tendency (purchase clusters 1 to N). For example, it is expected that movement line shape cluster 1 will be a group of movement lines moving clockwise in the store, 2 will be a group of movement lines moving counterclockwise, etc. Regarding the purchasing clusters, it is expected that the classification will be such that purchasing cluster 1 is a group of movement lines that purchase tea-related products individually, purchasing cluster 2 is a group of movement lines that purchase staple foods, etc. A movement line that is clockwise and purchases tea-related products individually will belong to movement line shape cluster 1 and purchasing cluster 1.

In addition, when clustering purchasing trends, clustering by number of items is possible. When clustering purchases, purchasing information linked to movement lines may be obtained from information DB 102, and clustering may be performed by number of items in the obtained purchasing information. This creates clusters by number of items purchased, such as a purchasing cluster for a single item purchase and a purchasing cluster for a two-item purchase, making it easier to grasp purchasing trends more clearly.

The cluster display unit 412 arranges clusters in a two-dimensional space based on the clustering results of the clustering unit 410, and presents them in association with purchasing information. First, the cluster display unit 412 classifies the movement lines by combining each cluster based on the movement line shape cluster and purchase cluster to which each movement line belongs, obtained from the clustering unit 410. Then, the movement lines classified into clusters are displayed on a diagram consisting of two axes, an axis representing the movement line shape cluster and an axis representing the purchase cluster.

13 is a diagram showing an example of an axis representing a flow line shape cluster and an axis representing a purchase cluster. The horizontal axis in FIG. 13 indicates each purchase cluster, and the vertical axis indicates the flow line shape cluster. Since each flow line belongs to each purchase cluster and flow line cluster, it is possible to classify and display based on the combination of clusters to which it belongs. Each circle in FIG. 13 indicates the number of flow lines belonging to each flow line shape cluster (1 to M) and purchase cluster (1 to N). Purchase information for purchase cluster 1 (tea), purchase cluster 2 (staple foods), and purchase cluster 3 (carbonated drinks) is shown. Focusing on the circle in (a), it represents a group of flow lines belonging to flow line shape cluster 3 and purchase cluster 1. Also, as in the first embodiment, the number of orders and the purchase unit price linked to the flow lines belonging to the combination of clusters may be represented by the size and color of the circle, respectively. In this way, the clustering unit 410 and the cluster display unit 412 can group multiple similar flow lines together, and further, the tendency of the flow lines can be understood from both the flow line shape and the purchase tendency.

FIG. 14 is a flowchart showing the flow of analysis processing by the analysis device 100 of the fourth embodiment. In the fourth embodiment, step S400 is executed after step S102.

In step S400, the CPU 11, as the clustering unit 410, classifies a plurality of similar movement paths based on the similarity between the received movement paths according to a predetermined clustering method, and clusters the movement paths.

In step S402, the CPU 11, as the cluster display unit 412, arranges clusters in a two-dimensional space based on the clustering results and presents them in association with purchasing information.

As described above, the analysis device 100 of the fourth embodiment can present multiple similar movements together in an analysis that takes into account the characteristics of customer purchases and movements, making it easy for the user to confirm.

[Fifth embodiment]
The fifth embodiment is an aspect in which a purchasing tendency is calculated using a product vector. The configuration of the fifth embodiment can be the same as that of the first embodiment (or the second to fourth embodiments).

The similarity calculation unit 110 of the fifth embodiment calculates the product vector based on the product's tendency to be sold together with other products obtained from the purchase information. The similarity calculation unit 110 calculates the purchasing tendency of each purchase by taking the sum of the product vectors of the products included in each purchase.

By generating vectors using the method shown in this embodiment, it is possible to capture purchasing trends based on how products are purchased. In this embodiment, product vectors are first generated based on the characteristics of other products that are likely to be sold together with each product, and purchase vectors are generated by combining the product vectors. For example, if "bread" has been sold together with "tea" once and with "milk" five times in the past, the bread vector will be "0, 1, 5". Each component of the vector represents the number of pieces of bread, tea, and milk that have been sold together with bread in the past. At this time, the vector may be normalized so that its norm is 1.

Furthermore, in retail stores, many products are purchased individually, and in order to represent such product features, a component of "number of times purchased individually" may be included in the vector. In this way, products that are sold alongside similar products are represented by similar vectors, making it possible to give them a higher meaning than existing product categories, such as "staple food" or "side dish." Based on the product vectors generated for each product in this way, a purchase vector is generated by adding up the product vectors of products included in a purchase. In the fifth embodiment, the similarity calculation unit 110 uses the purchase vectors calculated as above to perform the above-mentioned (c): distance based on purchasing tendency calculation.

As described above, the analysis device 100 of the fifth embodiment can calculate purchasing trends that reflect the tendency of products to be sold together, and can use this to calculate the similarity between traffic patterns.

Sixth Embodiment
The sixth embodiment is an aspect in which the strength of the relationship between the flow line shape and the purchasing tendency is evaluated using the clustering result. The configuration of the sixth embodiment can be the same as that of the fourth embodiment.

In the sixth embodiment, the clustering unit 410 uses the clustering results from the fourth embodiment to evaluate the strength of the relationship between the shape of traffic flow and purchasing tendency using mutual information as an index, and optimizes the classification of traffic flow into clusters and the clustering parameters (number of clusters) based on the evaluation value. Mutual information is a measure of the interdependence of two variables. By classifying traffic flow shapes and purchasing tendency as variables into clusters to maximize this index, and presenting the results of the cluster classification, it is expected that the relationship between the two can be shown more clearly.

For example, a method for optimizing the classification of movement lines into each cluster will be described. Consider a case where the number of movement lines is Nd, and the Nd movement lines are classified into M clusters based on the movement line shape. It is assumed that the purchasing tendency has already been classified into N clusters. The movement line shape cluster to which a certain movement line belongs is a random variable X (x∈{1, 2, ..., M), and the purchasing tendency cluster is Y (y∈{1, 2, ..., N). The histogram obtained by dividing the Nd movement lines into M x N clusters is normalized and captured as a probability distribution. In this way, the probability distribution P(X) of the movement line shape X, the probability distribution P(Y) of the purchasing tendency Y, and their joint distribution P(X, Y) can be obtained, and further the mutual information I(X; Y) between X and Y can be obtained. The mutual information is calculated for each grouping pattern that classifies the Nd movement lines into M clusters, and the final classification is performed using the pattern with the highest mutual information.

Figure 15 is a diagram showing the division of mutual information between grouping patterns A and B. Figure 16 is a diagram comparing the histograms of grouping patterns A and B. Note that the figures in parentheses in Figure 16 represent the joint probability when the histograms are viewed as a probability distribution. Since the mutual information of grouping pattern A is 0.53 and the mutual information of grouping pattern B is 0.17, grouping pattern A is selected. This method makes it possible to perform cluster classification that clearly reveals the relationship between traffic flow patterns and purchasing trends.

In the above example, purchasing trends have already been classified into clusters, and optimization is performed on the classification of movement line trends, but this can be reversed. Alternatively, both may not be determined, and the optimal classification pattern may be searched for based on the mutual information value for all patterns of M and N.

Next, an example of optimizing the number of clusters M and N will be explained. Existing clustering is performed by changing the values of M and N, and the mutual information is calculated using the above method based on the clustering results. Within that, a point is found where increasing M and N does not significantly increase the mutual information, and the values of M and N at this point are set as the final values of those parameters. By doing the above, it is possible to clarify the relationship between movement patterns and purchasing trends while suppressing an increase in the number of clusters that the operator must check, reducing the burden on the operator.

As described above, the analysis device 100 of the sixth embodiment can reduce the number of clusters that the user needs to check.

In addition, the analysis process that the CPU reads and executes the software (program) in each of the above embodiments may be executed by various processors other than the CPU. Examples of processors in this case include PLDs (Programmable Logic Devices) whose circuit configuration can be changed after manufacture, such as FPGAs (Field-Programmable Gate Arrays), GPUs (Graphics Processing Units), and dedicated electric circuits that are processors having a circuit configuration designed specifically to execute specific processes, such as ASICs (Application Specific Integrated Circuits). In addition, the analysis process may be executed by one of these various processors, or by a combination of two or more processors of the same or different types (for example, multiple FPGAs, and a combination of a CPU and an FPGA, etc.). More specifically, the hardware structure of these various processors is an electrical circuit that combines circuit elements such as semiconductor devices.

In addition, in each of the above embodiments, the analysis program is described as being pre-stored (installed) in the storage 14, but this is not limiting. The program may be provided in a form stored in a non-transitory storage medium such as a CD-ROM (Compact Disk Read Only Memory), a DVD-ROM (Digital Versatile Disk Read Only Memory), or a USB (Universal Serial Bus) memory. The program may also be downloaded from an external device via a network.

The following notes are further provided with respect to the above embodiment.

(Additional Note 1)
Memory,
at least one processor coupled to the memory;
Including,
The processor,
receiving purchase information and flow line information for each customer, and calculating a similarity between flow lines for each combination of flow lines based on a flow line shape obtained from the flow line information and a purchasing tendency obtained from the purchase information;
The analytical device is configured to:

(Additional Note 2)
A non-transitory storage medium storing a program executable by a computer to perform an analysis process,
receiving purchase information and flow line information for each customer, and calculating a similarity between flow lines for each combination of flow lines based on a flow line shape obtained from the flow line information and a purchasing tendency obtained from the purchase information;
Non-transitory storage media.

100 Analyzer 102 Information DB
110 Similarity calculation unit 112 Display unit 210 Flow line shaping unit 310 Extraction unit 410 Clustering unit 412 Cluster display unit

Claims (8)

a similarity calculation unit that receives purchase information and flow line information for each customer, and calculates a similarity between flow lines for each combination of flow lines based on a flow line shape obtained from the flow line information and a purchase tendency obtained from the purchase information;
An analytical device comprising: The analysis device according to claim 1, further comprising a display unit that displays each of the movement lines in a two-dimensional layout based on the similarity between the movement lines, so that the similarity between the movement lines is represented as a two-dimensional distance. The analysis device according to claim 2, wherein the display unit presents the purchasing information in association with each of the traffic lines arranged on the two-dimensional plane. The analysis device according to claim 2, wherein the display unit arranges each of the traffic lines on the two-dimensional plane with a size corresponding to a predetermined numerical value included in the purchasing information. The analysis device according to claim 1, further comprising a flow line shaping unit that divides a series of flow line route information in the flow line information into area units and converts the information into shaped route information on the area. The analysis device according to claim 1, further comprising an extraction unit that extracts information about a representative path of a traffic line from information about the path of a traffic line included in the traffic line information. The analysis device according to claim 1 calculates a product vector using the product and the concurrent sales tendency of other products of the product obtained from the purchase information, and calculates the purchase tendency of each purchase by taking the sum of the calculated product vectors. The analysis device according to claim 1, further comprising a clustering unit that classifies a plurality of similar movement paths according to a predetermined clustering method based on the similarity between the movement paths, and clusters the movement paths.