JP2002073943A - Method and device for classifying event transition pattern and storage medium storing event transition pattern classification program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method and device for classifying event transition patterns capable of processing at high speed with a relatively small storage capacity in order to classify clients for each event transition pattern from the time-series data of events that occur following the action of the clients, and a storage medium storing an event transition pattern classification program. SOLUTION: An event node serving as a record consisting of the names of events, a list of clients, the number of clients, and a list of pointers indicating the next event record is set, and the time-series data of the clients which comprise time, client identifiers, and a line of sets of event names are read. Based on the time-series data, the clients are classified for each event transition pattern and retained as event nodes at the time when all the time-series data are read, and as an event tree which shows a relationship of transition between the event nodes. The clients classified according to the event tree are output, together with the transition patterns.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an event transition pattern classification method and apparatus, and a storage medium storing an event transition pattern classification program. The present invention also relates to an event transition pattern classification method and apparatus for classifying clients according to event transition patterns, and a storage medium storing an event transition pattern classification program.

More specifically, effective marketing is performed by collectively extracting clients having similar movement patterns from log data in which positional information of clients moving while holding a portable terminal is recorded in chronological order. Support system for online shopping
An event transition pattern classification method in a support system for effective marketing by extracting clients having similar page browsing patterns and purchasing patterns from log data in which client access pages are recorded in chronological order, and The present invention relates to a device and a storage medium storing an event transition pattern classification program.

[0003]

2. Description of the Related Art Conventionally, as a method for extracting a transition pattern of a client, after listing the transition patterns for each client, the results are rearranged to classify the clients into a set having the same transition pattern. There is technology.

[0004]

However, in the above-mentioned conventional method, the storage capacity for holding the transition pattern for each client is large, and the processing time for rearranging is long. However, the decrease in execution speed and the increase in storage capacity become remarkable, which is not appropriate for handling large-size log data.

[0005] The present invention has been made in view of the above points, and a relatively small amount of storage is required to classify clients for each event transition pattern from time-series data of events that accompany client behavior. It is an object of the present invention to provide an event transition pattern classification method and apparatus capable of processing at high speed with a large capacity and a storage medium storing an event transition pattern classification program.

[0006]

FIG. 1 is a diagram for explaining the principle of the present invention.

The present invention (Claim 1) provides an event transition pattern classification method for classifying clients by event transition pattern based on time-series data of events that accompany client behavior. An event node consisting of a list of clients, the number of clients, and a list of pointers to the next event record is set (step 1), and the time series data of the client is read, which is a column of a set of time, client identifier, and event name (step 1). 2), clients are classified for each event transition pattern based on the time-series data, and the event nodes at the time when all of the time-series data are read are stored as an event tree representing a transition relation between the event nodes (step 3). ), Minutes based on the event tree And outputs the client that is (Step 4).

According to the present invention (claim 2), when a client is classified for each event transition pattern, a client data set consisting of different client identifiers appearing in the already read time-series data is searched based on the time-series data. If the client identifier of the time series data does not exist in the client data set, the client identifier is added to the list of clients of the event node, and the number of clients of the event node is added by 1,
When the client pointer of the client data set of the event node is set to point to the initially set event node, and the client identifier of the time-series data exists in the client data set, the number of clients of the event node Is subtracted by 1, the client identifier of the event node is deleted, and if there is the same event name as the time-series data in the list of pointers pointing to the next event record of the event node,
The client identifier of the time-series data is added to the client identifier of the event node of the event name, the number of clients is added by 1, the pointer pointed to by the client identifier is set as a pointer of a new event node, and the client of the time-series data is added. If the identifier does not exist in the client data set, an event node with the event name of the time-series data is newly generated, and the client identifier of the time-series data is added to the client identifier of the event node. Assuming that the number of clients of the node is 1, a pointer to a new event node is added to a list of pointers to the next event record of the event node, and the pointer of the client data set is updated with the pointer indicated by the client identifier.

According to the present invention (claim 3), when a client classified based on an event tree is output, a pointer of an event node to be output is initialized, a number indicating a transition pattern of the event node, The identifier and the number of clients for each transition pattern are output.

FIG. 2 is a diagram showing the principle of the present invention.

According to a fourth aspect of the present invention, there is provided an event transition pattern classifying apparatus for classifying a client according to an event transition pattern on the basis of time-series data of an event which accompanies an action of the client. Event node input means 150 for inputting an event node consisting of a name, a list of clients, the number of clients, and a list of pointers to the next event record;
A log data reading unit 110 for reading time-series data of a client including a series of a set of a time, a client identifier, and an event name; classifying clients for each event transition pattern based on the time-series data; and reading all the time-series data It has an event node at a certain point in time, an event transition pattern classifying means 120 for holding as an event tree representing a transition relationship between the event nodes, and an event transition pattern output means 130 for outputting a client classified based on the event tree. .

According to the present invention (claim 5), the event transition pattern classifying means 120 retrieves a client data set composed of different client identifiers appearing in the already read time-series data based on the time-series data. If the client identifier of the sequence data does not exist in the client data set, the client identifier is added to the list of clients of the event node, and the number of clients of the event node is added by 1,
Means for setting the client pointer of the client data set of the event node to point to the initially set event node; and, if the client identifier of the time-series data exists in the client data set, The number of clients is decremented by 1, the client identifier of the event node is deleted, and if there is the same event name as the time-series data in the list of pointers pointing to the next event record of the event node, the event is deleted. Means for adding the client identifier of the time-series data to the client identifier of the event node having the same name, adding 1 to the number of clients, and using a pointer indicated by the client identifier as a pointer of a new event node; If the client identifier is the client If that does not exist in the over data sets,
A new event node having the event name of the time-series data is newly generated, and the client identifier of the time-series data is added to the client identifier of the event node. Means for adding a pointer to a new event node to a list of pointers to event records, and updating the pointer of the client data set with the pointer indicated by the client identifier.

According to the present invention (claim 6), in the event transition pattern output means 130, a means for initializing a pointer of an event node to be output, a number indicating a transition pattern of the event node, a client identifier, and the transition Means for outputting the number of clients for each pattern.

According to a seventh aspect of the present invention, there is provided a storage medium storing an event transition pattern classifying program for classifying a client according to an event transition pattern based on time-series data of an event that accompanies a client action. An event node input process for inputting an event node consisting of an event name, a list of clients, a number of clients, and a list of pointers to the next event record, and a client consisting of a column of a set of a time, a client identifier, and an event name A log data reading process for reading time-series data, and classifying a client for each event transition pattern based on the time-series data, and an event node at the time when all the time-series data is read, and a transition relationship between the event nodes. Event tree Having an event transition pattern classification process holding, an event transition pattern output process for outputting the client classified based on the event tree as.

According to the present invention (claim 8), in the event transition pattern classification process, a process of searching for a client data set composed of different client identifiers appearing in the already read time-series data based on the time-series data; If the client identifier of the data does not exist in the client data set, the client identifier is added to the list of clients of the event node, the number of clients of the event node is incremented by 1, and the client data set of the event node is The process of setting the client pointer to point to the initially set event node, and, if the client identifier of the time-series data is present in the client data set, subtracting 1 from the number of clients of the event node, Event no Remove the client identifier,
In the list of pointers pointing to the next event record of the event node, if there is the same event name as the time-series data, the client identifier of the time-series data is set to the client identifier of the event node having the event name. In addition to the process of adding and adding 1 to the number of clients and using the pointer pointed to by the client identifier as a pointer of a new event node, if the client identifier of the time-series data does not exist in the client data set, Create a new event node of the event name of the series data, add the client identifier of the time series data to the client identifier of the event node,
The number of clients of the event node is set to 1, a pointer to a new event node is added to a list of pointers to the next event record of the event node, and the pointer of the client data set is updated with the pointer indicated by the client identifier. Having a process.

According to the present invention (claim 9), in the event transition pattern output process, a process of initializing a pointer of an event node to be output, a number indicating a transition pattern of the event node, a client identifier,
And a process of outputting the number of clients for each transition pattern.

As described above, in the present invention, the transition pattern is stored in a reduced form as an event tree without storing the transition pattern for each client, so that the storage capacity can be significantly reduced. Becomes

Further, in the present invention, if the event tree is searched only once for each node by the depth-first search, the correspondence between all transition patterns and their client sets can be output, so that the event sequence is rearranged as in the related art. There is no need, and the calculation speed can be greatly improved.

[0019]

FIG. 3 shows the configuration of an event transition pattern classification device according to the present invention.

The event transition pattern classification device 100 shown in FIG. 1 includes a log data reading unit 110, an event transition pattern classification unit 120, and an event transition pattern output unit 1.
30 and a client data set 140.

The log data reading unit 110 reads log data including a time, a client identifier, and an event.

The event transition pattern classifying unit 120 searches the client data set 140 based on the read log data, and classifies the client set by event transition pattern from the read log data.

The event transition pattern output unit 130 outputs, for each event transition pattern, a set of client sets that take that pattern.

The client data set 140 is a set of different client identifiers that have appeared in a portion that has already been read for given log data, and each client element has a pointer to a record that holds the data. .

Next, the operation of the above configuration will be described.

FIG. 4 is a flowchart of the event transition pattern classification processing of the present invention.

First, terms used in the following description will be described.

The "event node" is a record (event_seq) holding the following fields.

First field: Event name (ev
ent) second field: list of clients third field: number of clients (number of elements in second field) fourth field: list of pointers to next event record Step 101) first First, the user of the event transition pattern classification device determines that the event node (event_se
Enter an event name that can be a starting point in (q_start).

Step 102) It is determined whether or not unprocessed data (time, client, event) exists in the log data. If the data exists, the process proceeds to step 103. If not, an output process described later is performed. Move to.

Step 103) The log data reading section 110 sets a new time, a client identifier, and an event name (time, client, and event) from the log data.
nt) is read.

Step 104) The client identifier (client) of the read log data is searched from the client data set 140, and if the client identifier exists in the client data set, the process proceeds to step 105; Step 108
Move to

Step 105) In the event node (event_seq) indicated by the client identifier (client) of the client data set 140, the number of clients in the third field of the event node (event_seq) is reduced by one, and the number of clients in the second field is reduced. Delete the identifier (client).

Step 106) Fourth event node
Event name is "even" in the list of fields
It is determined whether or not there is a "t". If it exists, the process proceeds to step 107, and if not, the process proceeds to step 111.

Step 107) An event node (event_seq_u) having "event" as the event name
pdate), a new client (client) is added to the second field, the number of clients in the third field is increased by 1, and the event pointed to by the client (client) in the client data set 140 is changed from event_seq to event_seq_
Update to update, and proceed to Step 102.

Step 108) In the client data set 140, the read client identifier (cli)
ent).

Step 109) If the read event name (event) is an event node (event_se
If it is the event name of (q_start), step 102
Then, the process proceeds to step 110.

Step 110) Event node (e
If the event name is “vent_seq_start”, an event node (event_seq_star)
At t), a client identifier (client) is added to the second field, the number of clients in the third field is increased by 1, and the client data set 140 is added.
The pointer pointed to by the client identifier (client) in the event node (event_seq_s
start), and the process proceeds to step 102.

Step 111) In Step 106, when there is no event node whose event name is “event” in the list of the fourth field of the event node, an event node (event_seq_new) having “event” as the event name is added. Create a new client identifier in the second field (client)
Is added, the number of clients in the third field is set to 1, the number of pointers to “event_seq_new” is added to “event_seq” in the fourth field, and the pointer indicated by the client identifier in the client data set 140 is changed to “event_seq”. "To" event_seq_new ", and step 1
Move to 02.

Next, the event transition pattern output unit 130
The output processing of the event transition pattern classification result in the above will be described.

FIG. 5 is a flowchart of an output process of the event transition pattern classification result of the present invention.

Step 201) The pointer (ptr) is set to the event node (event_seq_start) (sequence = φ).

Step 202) The event node indicated by the pointer (ptr) is set as event_seq. Sequence the event node: sequence: = sequ
ence + event. Here, the event is an event node (event_event_point) indicated by the pointer (ptr).
seq) is the event name.

Step 203) Event node (ev
ent_seq), it is determined whether or not the number of clients in the third field is 0.
The process moves to step 05, and if it is not 0, the process moves to step 204.

Step 204) sequence, e
clie in the second field of vent_seq
nt and the number of clients in the third field are output.

Step 205) Event node (ev
ent_seq), it is determined whether there is any unsearched event node included in the fourth field, and if so, the process proceeds to step 206;
Move to 7.

Step 206) The pointer (ptr) is moved to an unsearched event node, and the routine goes to Step 202.

Step 207) Event node (ev
If there is an ancestor node whose unsearched event node is included in the fourth field of (ent_seq), the process proceeds to step 208; otherwise, the process ends.

Step 208) Move the pointer (ptr) to the ancestor node, delete the event name included in the event node passed from "sequence",
Select one unsearched event node and select a pointer (pt
r) to that event node.

[0050]

Embodiments of the present invention will be described below with reference to the drawings.

The following embodiment will be described with reference to the flowcharts of FIGS. 4 and 5 described above.

First, the event transition pattern classification processing will be described.

FIG. 6 shows an example of log data according to an embodiment of the present invention. In the following description, it is assumed that the log data is read.

In step 101, the event of the event node (event_seq_start) is set to e1
Set as

In step 103, (t1, c1, e1) on the first line is read from the log data, and
At 04, since there is no client c1, the process proceeds to step 108.

In step 108, c1 is added to the client data set 140. In step 110, the event node (event_seq_star) is added.
c1 is added to the second field of t), the third field is updated to 1, and c1 of the client data set 140 is updated.
1 is the event node (event_seq_star
Set to point to t), then go to step 102.

Again, the second line from the log data (t2, c2
, E1) are read and processed as described above. Also,
The third line (t3, c3, e1) is read and processed similarly. At this time, the client {c1, c2, c3} exists in the client data set 140, and the second node of the event node (event_seq_start) is present.
Is {c1, c2, c3}, and the third field is "3".

Next, (t4, c) in the fourth line from the log data
2, e3) are read. Client data set 140
, Since c2 exists, the process proceeds to step 105. In step 105, the event node (event
_Seq_start) is updated from “3” to “2”, and c2 in the second field is deleted.

In step 106, since there is no event whose event name is e3, the process proceeds to step 111, where an event node (event_seq_e3) is generated as event_seq_new. Further, the event name of event_seq_e3 is set to e3, c2 is added to the second field, and the third field is set to 1. Also, eve
ev is added to the fourth field of nt_seq_start.
A pointer to nt_seq_e3 is added, and the event node pointed to by c2 in the client data set 140 is set to event_seq_e3.

Next, (t5, c) in the fifth line from the log data
4, e 1) and (t 6, c 4, e 2) in the sixth row,
Each is processed in the same manner as the above procedure. At this point, the client data set 140 includes the client {c1
, C2, c3} exist and the event node (even
t_seq_start) is $ c
1, c3}, and the third field is "2". In addition, an event node (event_seq_e
The second field of 3) is {c2}, and the third field is "1".

The seventh line from the log data (t7, c3, e)
3) Read. In the client data set 140,
Since c3 exists, the process proceeds to step 105. In step 105, the event node (event_seq
_Start) from "2" to "1"
And delete "c3" in the second field. In step 106, since there is an event whose event name is "e3" in the fourth field, the flow proceeds to step 107.

In step 107, event_seq
_Update is an event node (event_seq
_E3), add “c3” to the second field,
Update the third field from "1" to "2". Finally, the event node indicated by "c3" in the client data set 140 is set to event_seq_e3. After the process is completed, the process proceeds to step 102, but since there is no data in the log data, the process proceeds to an output process described later.

Next, an output process of an event transition pattern will be described.

First, in step 201, the pointer (ptr) is set to the event node event_seq_s
set to start and sequence = φ. In step 202, sequence: = {e1}
Is set. In step 203, since the number of clients is 1, the process proceeds to step 204.

At step 204, "e1" is output as the sequence, {c1} of the second field is output, and the routine goes to step 205.

In step 205, an unsearched event node having an event name “e 2” is selected.
Shift to 07. In step 207, since there is no unsearched node, the pointer (ptr) is set to the event_seq
_Start, and the sequence: = {e1}
Update to Further, an unsearched event node having the event name "e3" is selected, the pointer ptr is moved to this event node, and the routine proceeds to step 202.

In step 202, the sequence
e: = {e1, e3} is set. Step 203
In step 204, since the number of clients is 2,
Move to In step 204, {e1, e3} is output as the sequence, and {c
2, c3} and the routine goes to step 205.

In step 205, since there is no unsearched node, the pointer is moved to event_seq_start and updated to sequence: = {e1}. Here, too, there is no unsearched node, so the processing is terminated.

An event tree at the time when all the log data is read and a tree representing a transition relation between them are called an event tree. FIG. 7 shows the event tree.

FIG. 8 is an output example of the event transition pattern classification according to the above-described procedure. With this, a set of all event transition patterns appearing in the log data and a set of clients taking the pattern can be listed.

The components of the event transition pattern classification device in the above embodiment are constructed as a program, and a disk device connected to a computer used as the event transition pattern classification device, a floppy (registered trademark) disk, a CD -The present invention can be easily realized by storing it in a portable storage medium such as a ROM and installing the same when implementing the present invention.

The present invention is not limited to the above embodiment, but can be variously modified and applied within the scope of the claims.

[0073]

As described above, according to the present invention, a relatively small amount of data is used to classify clients for each event transition pattern from time-series data (log data) of events that accompany client behavior. High-speed processing can be performed with a storage capacity.

[Brief description of the drawings]

FIG. 1 is a diagram for explaining the principle of the present invention.

FIG. 2 is a principle configuration diagram of the present invention.

FIG. 3 is a configuration diagram of an event transition pattern classification device of the present invention.

FIG. 4 is a flowchart of an event transition pattern classification process of the present invention.

FIG. 5 is a flowchart of an event transition pattern classification result output process according to the present invention.

FIG. 6 is an example of log data according to an embodiment of the present invention.

FIG. 7 is a diagram illustrating an event tree according to an embodiment of the present invention.

FIG. 8 is an example of an output result of one embodiment of the present invention.

[Explanation of symbols]

REFERENCE SIGNS LIST 100 event transition pattern classification device 110 log data reading means, log data reading section 120 event transition pattern classification means, event transition pattern classification section 130 event transition pattern output means, event transition pattern output section 140 client data set 150 event node input means

Claims (9)

[Claims] 1. An event transition pattern classification method for classifying clients according to event transition patterns based on time-series data of events that accompany client behavior, comprising: an event name, a list of clients, the number of clients,
Set an event node that is a record consisting of a list of pointers to the next event record, read the client's time-series data consisting of a set of time, client identifier, and event name columns, and transition events based on the time-series data. Clients are classified for each pattern, and the event nodes at the time of reading all the time-series data and an event tree representing a transition relationship between the event nodes are held as an event tree. The clients classified based on the event tree are transition patterns. And an event transition pattern classification method. 2. When classifying a client for each of the event transition patterns, a client data set including different client identifiers appearing in the already read time-series data is searched based on the time-series data. If the client identifier of the event node is not present in the client data set, the client identifier is added to the list of clients of the event node, the number of clients of the event node is incremented by 1, and the client data set of the event node is added. Set the client pointer of the client to point to the initially set event node, and if the client identifier of the time-series data is present in the client data set, subtract 1 from the number of clients of the event node; The event node The client identifier is deleted, and if a list of pointers to the next event record of the event node has the same event name as that of the time series data, the client identifier of the event node having the event name has the time series data. The client identifier of the data is added, and the number of clients is incremented by 1. The pointer pointed to by the client identifier is used as a pointer of a new event node. When the client identifier of the time-series data does not exist in the client data set, Generates a new event node having the event name of the time-series data, adds the client identifier of the time-series data to the client identifier of the event node, sets the number of clients of the event node to 1, and Returns a pointer to the next event record. 2. The event transition pattern classification method according to claim 1, wherein a pointer to a new event node is added to the list, and the pointer of the client data set is updated with the pointer indicated by the client identifier. 3. When outputting a client classified based on the event tree, a pointer of an event node to be output is initialized, a number indicating a transition pattern of the event node, a client identifier, and the transition. 2. The event transition pattern classification method according to claim 1, wherein the number of clients for each pattern is output. 4. An event transition pattern classifying apparatus for classifying a client according to an event transition pattern based on time-series data of an event generated in association with an action of the client, comprising: an event name; a list of clients; number,
Event node input means for setting an event node which is a record comprising a list of pointers pointing to the next event record; log data reading means for reading client time-series data consisting of a set of time, client identifier, and event name Classifying a client for each event transition pattern based on the time-series data, and storing an event node at the time when all the time-series data is read, and an event transition pattern classification for retaining an event tree representing a transition relation between the event nodes; And an event transition pattern output unit that outputs a client classified based on the event tree together with a transition pattern. 5. The event transition pattern classification means, comprising: means for searching, based on the time-series data, a client data set consisting of different client identifiers appearing in the already-read time-series data; Is not present in the client data set, the client identifier is added to the list of clients of the event node, the number of clients of the event node is incremented by 1, and the client of the client data set of the event node is added. Means for setting a pointer to point to the initially set event node; and if the client identifier of the time-series data is present in the client data set, subtract 1 from the number of clients of the event node. Event node client If the client identifier of the event node of the event name is deleted from the list of pointers pointing to the next event record of the event node, the client identifier of the event node of the event name is deleted. Means for adding the client identifier of the data, adding 1 to the number of clients, and using a pointer indicated by the client identifier as a pointer of a new event node, wherein the client identifier of the time-series data exists in the client data set If not, a new event node of the event name of the time series data is newly generated, the client identifier of the time series data is added to the client identifier of the event node, the number of clients of the event node is set to 1, and Pointer to the next event record of the event node 5. The event transition pattern classification device according to claim 4, further comprising: a means for adding a pointer to a new event node to the list of the client data set and updating the pointer of the client data set with the pointer indicated by the client identifier. 6. The event transition pattern output means: means for initializing a pointer of an event node to be output, a number indicating a transition pattern of the event node, a client identifier, and the number of clients for each transition pattern. 5. The event transition pattern classification device according to claim 4, further comprising: means for outputting the event transition pattern. 7. A storage medium storing an event transition pattern classification program for classifying clients according to event transition patterns based on time-series data of events that accompany client actions, comprising: an event name; List of clients, number of clients,
An event node input process for setting an event node, which is a record consisting of a list of pointers to the next event record; a log data reading process for reading a client's time-series data consisting of a set of time, client identifier, and event name Classifying a client for each event transition pattern based on the time-series data, and storing an event node at the time when all the time-series data is read, and an event transition pattern classification for retaining an event tree representing a transition relation between the event nodes; A storage medium storing an event transition pattern classification program, comprising: a process; and an event transition pattern output process for outputting a client classified based on the event tree together with a transition pattern. 8. The event transition pattern classification process includes: searching a client data set composed of different client identifiers appearing in the already read time-series data based on the time-series data; Is not present in the client data set, the client identifier is added to the list of clients of the event node, the number of clients of the event node is incremented by 1, and the client of the client data set of the event node is added. A process of setting a pointer to point to the initially set event node; and, if the client identifier of the time-series data exists in the client data set, subtract 1 from the number of clients of the event node. Event If the client identifier of the event node of the event node having the same event name as the time-series data in the list of pointers pointing to the next event record of the event node is deleted, the client identifier of the event node having the event name is deleted. A process of adding the client identifier of the time-series data, adding 1 to the number of clients, and setting a pointer pointed to by the client identifier as a pointer of a new event node; If it does not exist in the set, a new event node with the event name of the time-series data is newly generated, the client identifier of the time-series data is added to the client identifier of the event node, and the number of clients of the event node is set to 1 And the next event record of the event node A process of adding a pointer to a new event node to a list of pointers pointing to the client data set and updating the pointer of the client data set with the pointer pointed to by the client identifier. Storage media. 9. The event transition pattern output process includes a process of initializing a pointer of an event node to be output, a number indicating a transition pattern of the event node, a client identifier, and the number of clients for each transition pattern. And outputting the event transition pattern classification program according to claim 7.