JP2001306998A - Time series analysis method - Google Patents

Abstract

(57) [Summary] [Problem] To clarify a causal relationship by comparing a change in event occurrence data on a time series with an event at the time of occurrence, and to predict a change in data based on the relationship. The present invention relates to a method for analyzing a time series of events. The analysis data collection conditions and analysis conditions are set, and event data to be analyzed is collected from past accumulated data in accordance with the collection conditions, and sorted in chronological order by analysis target item. The sorted item-by-item event data is compared with the model pattern arranged in time series, and when the item-by-item event data matches the model pattern within a predetermined error range, the item-by-item event data is extracted as a feature pattern, The pattern of the event data extracted as the feature pattern is compared with a previously prepared event calendar, and when it is determined that there is a correlation between the change in the event data and the event calendar, the event and timing caused by the data change And visualizes the found event data together with the occurrence time data and outputs it.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a time series analysis method for analyzing a time series of data handled by a system for collecting and managing a large amount of event data generated in a time series such as a POS system (point of sale system). About.

[0002]

2. Description of the Related Art In a point-of-sale system or the like, when the sales data of a specific product is viewed in a time series, the sales amount is rarely substantially constant throughout the year, and many products have a slight fluctuation. A certain product has a sharp increase in sales at a specific time, and a certain product is excluded at a specific time. Time-series analysis is useful for making a product plan based on such time-series fluctuation.

[0003] If there is a significant change in the sales volume, it can be a great force in product planning if it is known what kind of event it is affected by. In the past, when estimating the cause of such a change, it was largely determined by experienced persons. For example, "Sell a cake at Christmas" is something everyone knows, but "Sell a padlock in mid-April" is hard to understand unless you are someone with considerable work experience. This requires that you know as experience after the entrance ceremony that you need a locker lock and that will lead to consumption.

[0004]

SUMMARY OF THE INVENTION According to the present invention, even a person who has no knowledge or experience of an event or the amount of occurrence of the event can recognize the change in the time series of the event occurrence data and the event at the time of occurrence. The purpose of the present invention is to clarify a causal relationship by comparing the data, and to predict a change in data based on the causal relationship.

[0005] Also, for a large amount of time-series data,
It is an object of the present invention to provide a method for finding a relationship between a feature of data and an event that has not been noticed before.

[0006]

In order to achieve the above object, the invention according to claim 1 is directed to a method for collecting collection conditions for collecting data to be analyzed from past accumulated data and collecting collected event data. A setting step for setting analysis conditions for analysis, a collection step for collecting event data to be analyzed from past accumulated data in accordance with collection conditions, and sorting the items in the analysis target in chronological order, and a collection step for sorting. The itemized event data is compared with a model pattern prepared in advance and arranged in a time series. When the itemized event data matches the model pattern within a predetermined error range, the itemized event data is extracted as a feature pattern. The extraction step and comparing the pattern of the event data extracted as the feature pattern with a previously prepared event calendar,
When it is determined that there is a correlation between the change in the event data and the event calendar, an associating step of finding an event and a time due to the data change, and the event data determined to be correlated in the associating step. And an output step of visualizing and outputting the data together with the occurrence time data.

According to a second aspect of the invention, in the time series analysis method according to the first aspect, at least one of an event, an event, an event, and a climate is provided as an event pattern prepared in advance and arranged in a time series. It is characterized by including three patterns.

The invention according to claim 3 is the invention according to claim 1 or 2
Wherein the event data corresponds to the sales or the sales quantity of the product.

According to a fourth aspect of the present invention, a setting step for setting a collection condition for collecting data to be analyzed from past accumulated data and an analysis condition for analyzing collected event data, A collection step of collecting event data to be analyzed from the accumulated data according to the collection conditions and sorting the analysis target items in chronological order by item to be analyzed, and the event data by item sorted in the collection step are arranged in a time series prepared in advance. Compared with a model pattern that changes from a low value to a high value or from a high value to a low value and continues for a predetermined period or more, and when the item-specific event data matches the model pattern within a predetermined error range, the item-specific event data is An extraction step for extracting as a feature pattern, and associating the event data extracted as a feature pattern in the extraction step with a time on a calendar A response with steps and proposes a time series analysis method and an output step of visually of outputting the event data associated with mapping step as a function of time.

According to a fifth aspect of the present invention, in the time-series analysis method according to the fourth aspect, the event data corresponds to a sales amount or a sales quantity of a product.

[0011]

<Embodiment 1> The first to third aspects of the present invention will be described with reference to FIGS. This embodiment seeks to find data whose value changes rapidly when viewed in chronological order from the chronological data. As shown in the flowchart of FIG. Data collection step 20 for collecting data for analysis, extraction step 3 for extracting desired data
0, an associating step 40 for associating the extracted data with the event, and an output step 50 for outputting the associated data. Hereinafter, each step will be described in detail with reference to FIG.

In setting step 10, setting screen 1 shown in FIG.
As shown at 00, an analysis target table is designated, for example, as a "weekly sales table" from a database table (not shown). In addition to weekly sales tables, there may be various types of database tables of this type, such as monthly sales tables and daily sales tables. The collection conditions and the analysis conditions are set as follows for each field item included in the specified weekly sales table. That is, for setting items such as a time field, an analysis target field,..., NULL value processing, etc.
, The analysis target field is sales volume, the aggregation unit field is store and category 2, the maximum duration is "3",
The cutoff value of the degree of change is “3”, and the NULL value processing is “0”.
Is set. Note that the first week of the year-week may be set to the first week of January in the calendar year, but is not limited thereto. Set it as the first week of another month,
Can be set freely.

In the data collection step 20, as shown in FIG. 3, the weekly sales table 200 set as the target analysis table in the setting step 10 is referred to, and in accordance with the field setting contents set in the setting step 10, the store,
Mark the fields for Category 2, Year, Week, and Sales Volume, and refer only to the field items marked in FIG.
FIG. 4 shows sales quantities by store (for example, A, B,..., N) and by sub-category (for example, a, b,. , And 30n as shown in FIG.

In the extraction step 30, the classification tabulation sheets 301, 302,.
Are sales model patterns 401 and 40, which are prepared in advance as time series graphs (or time series lists), as shown in FIG.
2, 403,... To check whether or not the former matches the latter within a predetermined error range, and if they match, extract them as shown in FIG. Output as In the illustrated example, three pieces of time-series data are extracted as feature patterns corresponding to the matching condition.

Next, in the associating step 40, the time series data extracted in the extracting step 30 is compared with a previously prepared event calendar. 7 to 9 show the event calendar for June as an example. FIG. 7 shows a calendar 420 in which yearly events, events, events, and anniversaries in June are arranged in chronological order.
Until the day of “Kimono Day”, the list of existing publicly known affairs is listed. FIG. 8 is a calendar 43 that summarizes “life” habitual (lifestyle) events and concerns on a weekly basis.
0. From the first week, "Clothes and Living Change" to the fourth week, "Summer Festival of Prayer for Good Harvest" are summarized.
FIG. 9 is a calendar 440 showing a table in which meteorological data such as temperature, humidity, weather, and the like in the same period of the previous year (or the average of recent years) are filled in order of date (numerical values are omitted in the figure). In some cases, a discomfort index may be added in relation to temperature and humidity. These event calendars 420 to 440 shown in FIGS. 7 to 9 are used as appropriate. The change of the time-series data extracted in the extraction step 30 is compared and collated with the event calendars of FIGS. 7 to 9, and data associated with the correlation between the change of the time-series data and the event is determined. Here, it is determined, for example, that "slippers will be sold in June. → Influence of class visits !?"

Next, in the output step 50, as shown in FIG. 10, the time-series data determined to have a correlation in the associating step 40 is read as follows: "Slippers are sold in June. → Influence of class visit! "?" Or the like and output as an output sheet 450 or an output screen.

As described above, according to the first embodiment, temporarily changing data can be found from the time-series data.

<Embodiment 2> The invention according to claims 4 to 6 will be described with reference to FIGS. This embodiment finds data that changes rapidly from time-series data, and applies the time of the change to an annual calendar by pairing an ascending change and a descending change, thereby obtaining data indicating a periodical change. As shown in the flowchart of FIG. 11, a setting step 110 for setting analysis conditions, a data collecting step 120 for collecting data for analysis, and an extracting step 130 for extracting desired data are to be found. , An associating step 140 for associating the extracted data with the event, and an output step 150 for outputting the associated data. Hereinafter, each step will be described in detail with reference to FIGS.

Steps 110 and 120 are the same as steps 10 and 20 described above, respectively. In the extraction step 130, the classification tabulation sheets 301, 302,..., 30n obtained in the collection step 120 (see FIG. 4) are
As shown in 2, time series graph (or time series list)
Sales model patterns 501, 502, and 5 showing the model patterns of the sales quantity for the year and week prepared in advance as
, And whether or not they match within a predetermined error range. If they match, they are extracted as described above (see FIG. 6). Here, the sales model patterns 501, 502, 503,... Are viewed as time-series data. For example, the sales model patterns change from a low value to a high value, maintain a high value for a predetermined period or more, and then return to a low value. The other is a pattern that changes from a high value to a low value, maintains a low value for a predetermined period or more, and then returns to a high value. Moreover, here, the change in the data value is considered as a pair of data change from a low value to a high value and a change from a high value to a low value, and a change from a high value to a low value and a change from a low value to a high value. Is regarded as a pair of data changes.

The event data extracted in this way is
In the associating step 140, for each product classification item, for example, a sales period is associated on the calendar with contents as shown in FIG.

Finally, in output step 150, FIG.
As shown in FIG. 3, the value of the data is visualized in the form of a 600-sheet or screen of annual sales timing by product category in a form in which data values are expressed on a year-week calendar so that sales timings can be listed for each product classification item. Output. In the figure, products satisfying a predetermined condition are expressed in association with years and weeks. However, in some cases, events corresponding to each period can be described as reference data in years and weeks.

According to the second embodiment, data that changes with a certain period width can be found from the time-series data.

In the above-described embodiment, the events to be taken into the event calendar have been described by exemplifying an event, an event, an event, an anniversary, a lifestyle-related event or interest, weather data, and the like. By incorporating local events, television commercials, flyer distribution, etc., it is expected that more effective results will be derived.

[0024]

According to the present invention, it is possible to easily find the correlation between the characteristic change of the time series data and the events related to the time series.

Like "the cake can be sold at Christmas" that everyone considers "natural", or "slippers can be sold at the time of class visits" so that if you know the result, you can be convinced that "I see." In addition, unexpected and unexpected events can be correlated with timing.

[Brief description of the drawings]

FIG. 1 is a flowchart showing an embodiment of the invention according to claim 1;

FIG. 2 is a view showing a setting screen in a setting step of FIG. 1;

FIG. 3 is a chart showing a weekly sales table used in the embodiment shown in FIG. 1;

FIG. 4 is a table showing a classification tabulation sheet used in the embodiment shown in FIG. 1;

FIG. 5 is a view showing a sales model pattern used in the embodiment shown in FIG. 1;

FIG. 6 is a chart of an output sheet showing an output form of an analysis result in the embodiment shown in FIG. 1;

FIG. 7 is a chart showing calendars for events, events, events, and anniversaries used in the embodiment shown in FIG. 1;

FIG. 8 is a chart showing a life-related calendar used in the embodiment shown in FIG. 1;

FIG. 9 is a chart showing a calendar related to weather used in the embodiment shown in FIG. 1;

FIG. 10 is an explanatory diagram showing an output form in the embodiment shown in FIG. 1;

FIG. 11 is a flowchart showing an embodiment of the invention according to claim 4;

FIG. 12 is a view showing a sales model pattern used in the embodiment shown in FIG. 11;

FIG. 13 is an explanatory diagram showing an output form in the embodiment shown in FIG. 11;

[Explanation of symbols]

10, 110 setting step 20, 120 collecting step 30, 130 extracting step 40, 140 associating step 50, 150 output step 100 setting screen 200 weekly sales table 301 to 30n classification and aggregation sheet 401 to 403 sales model pattern 410 output sheet 420 Calendars for events, events, events, and anniversaries 430 Life-related calendars 440 Weather calendars 450 Output sheets 501-503 Sales model patterns 600 Annual sales timing list sheets by product category

Claims (5)

[Claims] A setting step for setting a collection condition for collecting data to be analyzed from past accumulated data and an analysis condition for analyzing collected event data; A collection step of collecting event data to be analyzed in accordance with the collection conditions and sorting the analysis target items in chronological order, and the event data by item sorted in the collection step,
An extraction step of comparing with the model patterns prepared in advance and arranged in a time series, and extracting the item-specific event data as a feature pattern when the item-specific event data matches the model pattern within a predetermined error range; The pattern of the event data extracted as the feature pattern is compared with a previously prepared event calendar, and when it is determined that there is a correlation between the change in the event data and the event calendar, the event caused by the change in the data A time-series analysis method comprising: an associating step of finding a time; and an output step of visualizing and outputting event data determined to have a correlation with the occurrence time data in the associating step. 2. The time series analysis method according to claim 1, wherein the event patterns prepared in advance and arranged in a time series include at least one pattern of an event, an event, an event, and a climate. . 3. The time-series analysis method according to claim 1, wherein the event data corresponds to sales or sales quantity of a product. 4. A setting step for setting a collection condition for collecting data to be analyzed from past accumulated data and an analysis condition for analyzing collected event data, and A collection step of collecting event data to be analyzed in accordance with the collection conditions and sorting the analysis target items in chronological order, and the event data by item sorted in the collection step,
Compared with a model pattern prepared in advance and arranged in time series, changing from a low value to a high value or from a high value to a low value and continuing for a predetermined period or more, the item-specific event data is within a predetermined error range for the model pattern. An extracting step of extracting the item-specific event data as a feature pattern when matching, an associating step of associating the event data extracted as a feature pattern in the extracting step with a time on a calendar; An output step of visualizing and outputting the obtained event data as a function of time. 5. The time-series analysis method according to claim 4, wherein the event data corresponds to a sales amount or a sales quantity of a product.