KR100902485B1 - Method for exctracting frequent access pattern of mobile user with time - Google Patents

Abstract

The present invention is a method of extracting a frequent behavior pattern of a mobile user according to a time for converting a behavior pattern of a mobile user's time zone into a tree structure and extracting a frequent tree path, and defining a time stamp in which a day is divided by a unit of time. In the first step, the location information and service information of each mobile user are integrated for each time zone defined in the time stamp, and each mobile user includes a user ID, a time stamp, and service information according to the location information and location information. The second step of generating integrated log data including time-phased behavioral patterns for and storing them in the integrated database, scans the integrated database, extracts the location information having support above the threshold value among the behavioral patterns of the integrated log data, and extracts the header table. Third step to generate, based on the header table, consolidated log data A fourth step of constructing a location information tree for the location information pattern included in the behavior pattern of the fifth step of constructing a service information tree for the service information pattern according to each location information pattern, and a fourth step and a fifth step And extracting a frequent tree path from the location information tree and the service information tree generated by the method.

Data mining; Ubiquitous; Behavior pattern

Description

How to extract frequent behavior patterns of mobile users over time {METHOD FOR EXCTRACTING FREQUENT ACCESS PATTERN OF MOBILE USER WITH TIME}

The present invention relates to a method of extracting frequent behavior patterns of a mobile user over time, and more particularly, after converting a behavior pattern of a mobile user into an easy tree structure for a semi-structured data format that is a communication between mobile agents. The present invention relates to a method of extracting frequent behavior patterns of mobile users, that is, useful behavior patterns, by time zone.

Ubiquitous computing provides an environment where users can receive useful services anytime, anywhere using wireless terminals such as PDAs or mobile phones that are ubiquitous in everyday life. In addition, in the ubiquitous environment, it is possible to extract service behavior patterns according to the user's location by using semi-structured data, that is, situational recognition data with hierarchical data structure, through intelligent multi-mobile agent communication.

However, in the conventional method of extracting user behavior patterns using only location information and service information, only the frequent behavior patterns of statistical users are extracted. Accordingly, there is a problem in that the user cannot actively cope with the dynamic change of the user, that is, the service request of the user over time.

With the combination of the visualization of the ubiquitous environment that enables information technology (IT) to be freely connected to the network anytime, anywhere, regardless of computer or network conditions, and the intelligent environment that frees users from the direct operation of computing as much as possible, Likewise, the extraction and access of information and knowledge by simple statistical analysis has become limited. Therefore, through the mining technique that can extract the temporal information of the efficient and semi-structured data in the intelligent ubiquitous environment, various application systems are urgently required along with the real-time user-oriented services that meet the intelligent ubiquitous environment.

The present invention has been proposed in order to solve the above-mentioned conventional problems, and an object of the present invention is the frequent change of the mobile user over time, which can actively cope with the dynamic change of the user, that is, the service request of the user over time. It is to provide a method of extracting behavior patterns.

In order to achieve the above object, the frequent behavior pattern extraction method of the mobile user according to the time according to the present invention, after converting the behavior pattern for each time zone of the mobile user into a tree structure according to the time to extract the frequent tree path A method of extracting a frequent behavior pattern of a mobile user, the first step of defining a time stamp that divides a day into a certain unit of time, by integrating the location information and service information of each mobile user for each time zone defined in the time stamp, A second step of generating integrated log data including a user ID, a time stamp, and a time-phase behavior pattern for each mobile user including location information and service information according to the location information, and storing the data in an integrated database; Scanning and positioning with support above the threshold among the behavior patterns of the consolidated log data A third step of generating a header table by extracting the data stream, a fourth step of constructing a location information tree for the location information pattern included in the behavior pattern of the integrated log data based on the header table, and a service according to each location information pattern And a sixth step of constructing a service information tree for the information pattern, and a sixth step of extracting a frequent tree path among the location information tree and the service information tree generated in the fourth and fifth steps. .

Preferably, the third step is a step 3-1 of scanning the unified database and counting each location information included in the behavior pattern of the unified log data irrespective of the order, and the location information having the counted information smaller than the threshold value. Include step 3-2 of removing. Further, the fourth step includes steps 4-1 for giving a null value to the root, and sequenced position information included in the behavior pattern of each time stamp for each user ID of the integrated log data to each node. And a fourth step of generating and connecting to each other, wherein each node includes a location information and a count of generating location information patterns corresponding to a tree path to each node. In this case, in step 4-2, when the location information included in the specific behavior pattern of the integrated log data is already generated in the node of the location information tree, the count of the existing node is increased instead of creating a new node. In addition, the fifth step is the fifth step of giving a null value to the root, the location information pattern included in each tree path of the location information tree of the service information included in the behavior pattern of the integrated log data Step 5-2 of configuring nodes with service information for and connecting them to each other, step 5-3 of connecting a root of each service information tree to a tail node of a tree path of a corresponding location information tree, and the level of each node And a fifth step of connecting each time stamp and each node of the service information tree header table including a time stamp, wherein each node of the service tree corresponds to a service path and a tree path to each node. Include counts in which information patterns have occurred. More preferably, the sixth step is a sixth step of pruning the tree path and the connected service tree when the count of each node of the tree path is less than the preset frequency with respect to the location information tree, and For the service information tree, if the count of each node of the tree path is less than the frequent frequency, step 6-2 of pruning the tree path is included.

As described above, the frequent behavior pattern extraction method of the mobile user over time according to the present invention can extract useful behavior pattern information over time of the mobile user from the semi-structured data in an intelligent ubiquitous environment using a mining technique. This has the remarkable effect of providing the foundation of various application systems as well as various user-oriented services.

The present invention extracts useful behavior patterns of mobile users using contextual awareness data consisting of semi-structured data that is communication between mobile agents in an intelligent ubiquitous environment. Specifically, the time stamp is pre-defined by the user and the administrator in the preprocessing process and then combined with the situation recognition data which is composed of semi-structured data. That is, after defining the time stamp in advance, the integrated log data is generated by integrating the location information and service information of the mobile user according to the time stamp. Consolidated log data is stored in the consolidated database. The integrated database is a tuple-based, limited in memory loading and quick access to mobile services. Therefore, after transforming into a tree structure, the user's location information and service information for each time zone determined by the user and the administrator must be extracted. This enables individual data management, preventing access restrictions and speed degradation, as well as easily converting into a more compact and easy-to-load tree structure. Also, by representing the location information and the service information of the user in a tree structure, the behavior pattern of the mobile user can be extracted more efficiently. In addition, by discovering frequent patterns of users by time zones, the advertisement planner and national public service advertisements can provide appropriate information for each mobile user at each time zone, and can provide appropriate real-time services to mobile users in a ubiquitous environment.

Hereinafter, with reference to the accompanying drawings will be described in detail the advantages, features and preferred embodiments of the present invention.

FIG. 1 is a diagram illustrating a system configuration in which an intelligent middleware platform and a mining module are integrated, and illustrates a state in which a mining module is linked to middleware. In the system of FIG. 1, the CALM platform was used as the middleware platform. As shown in FIG. 1, by integrating a mining module into an intelligent mobile agent of a middleware platform in a ubiquitous computing environment, the user's requirements and useful information can be effectively mined from the accumulated data. Behavior patterns can be extracted. In detail, the intelligent mobile agent accumulates the information of the mobile terminal through each sensor and effectively mines the accumulated information by using the mining module, and then may be linked with the decision system to assist the user in making a decision.

2 is a system configuration diagram for extracting a frequent behavior pattern of a mobile user over time according to the present invention, which shows a detailed system configuration of a mining module in FIG. As shown in FIG. 2, the intelligent mobile agent of the middleware stores time log information of a mobile terminal user in a time zone location information of the mobile terminal and service requested log information. Accumulate by classifying. Accumulated location information and service information are integrated in each time zone and then stored in an integrated database. The log data (integrated log data) of the time zone location information and the service information stored in the integrated database is used to extract user requirements and desired information by the mining algorithm according to the present invention.

On the other hand, the problem definition used to extract the useful behavior pattern of the mobile user over time in the present invention is the following equation (1).

In Equation 1, l denotes location information and s denotes service information, and (l: s) ^t to l: s denote an information set of an equal time zone (t) of location information (l) and service information (s). Indicates. Therefore, through the equation (1) it is possible to extract the behavior pattern of the mobile user for each time zone using the location information and service information, and to extract the useful frequent behavior pattern of the user for each time zone.

The user's useful behavior pattern, that is, the support for the user's frequent behavior pattern may be expressed as Equation 2.

Service information is extracted from the integrated log data of the entire integrated database in each time zone, and it is determined to be frequent when the user or administrator exceeds a predetermined threshold.

3 is a flowchart of a process of extracting a frequent behavior pattern of a mobile user over time according to the present invention. Referring to Figure 3 describes the process of extracting the frequent behavior pattern of the mobile user over time according to the present invention.

First, before performing a mining process for extracting a behavior pattern of a mobile user over time, a time stamp is defined as a preprocessing process (ST100). The time stamp is for extracting desired information for each desired time zone and may be defined by dividing the day into a predetermined time unit.

When the time stamp is defined, one integrated log data is generated by integrating location information and service information for each defined time zone (ST110). At this time, the integrated log data is composed of a user ID, a time stamp and a behavior pattern, and the behavior pattern is composed of location information and service information accordingly. The service information gives a value preset by an administrator or a user, and has sequenced information. Integrated log data is stored in the integrated database. Thereafter, a process of converting the integrated log data stored in the integrated database into a compact tree structure is performed.

In order to convert the integrated log data into a tree structure, first, a header table is extracted from each behavior pattern of the integrated log data (ST120). The header table is generated by extracting location information having a support level greater than or equal to a threshold value among behavior patterns of the integrated log data. Specifically, after scanning the unified database to count each location information included in the behavior pattern of the unified log data, only the location information having a count equal to or greater than the threshold value is extracted. The location information having a count less than the threshold is non-frequent location information, and in the non-frequent location information, the service information is also rare and thus removed. At this time, the threshold is preset by an administrator or a user.

When the header table is generated in ST120, a tree for the location information is constructed based on this (ST130). The tree for the location information is created by assigning a null value to the root, and then displaying and connecting the location information included in the behavior pattern to each node by time stamp of each user ID. That is, since the location information of the behavior pattern for each user ID of each user ID is sequenced with each other, the tree for the location information can be constructed by configuring these location information as nodes and connecting them to the root. At this time, each node displays a count of occurrence of a pattern of the corresponding location information together with the location information. In addition, when the location information included in the specific behavior pattern is already generated as a node in the location information tree, only the count is raised without creating a new node. At this time, location information not included in the header table is not represented in the tree due to having a support degree smaller than the threshold. That is, the tree for the location information consists only of the location information included in the header table.

When the tree for the location information is completed, a tree for the service information is constructed (ST140). The tree for the service information is also created by assigning a null value to the root, and then constructing and connecting nodes with the service information included in the behavior pattern of the integrated log data. Specifically, the service information for each pattern of location information of the location information tree is configured and connected to each node. At this time, each node of the service information tree also displays the count of occurrence of the corresponding service information pattern together with the service information. When the service information tree is generated, the root of each service information tree is connected to the tail node of the corresponding location information tree. That is, the root of the service information tree is connected to the tail node of the tree of the location information pattern for the service information pattern of each service information tree.

When the location information tree and the service information tree are completed, frequent paths are extracted from the generated tree (ST150). In this case, a criterion for extracting a frequent path may be to use a frequent degree preset by an administrator or a user, and may use a value equal to a threshold used for generating a header table. First, in the location information tree and the service information tree, the degree of frequent occurrence of each location information included in the location information pattern, that is, the count of each node of the location information tree path is removed by pruning the tree path whose frequency is smaller than the frequency. At this time, if the tree path of the location information is pruned, the corresponding service information tree is also pruned. Next, the tree paths whose frequency of each service information included in each service information pattern of the service information tree, i.e., the count of each node of the service information tree path is less than the frequent frequency, are pruned. After pruning the infrequent tree paths, only the frequent tree paths with minimum support are left. Therefore, through the frequent tree path, it is possible to extract all the frequent, useful behavior patterns, by time zone (by time stamp) of the mobile user.

Hereinafter, a process of extracting a useful behavior pattern for each time zone of a mobile user will be described in more detail, for example.

4 is an exemplary view of a time stamp according to the present invention. The time stamp shown in FIG. 4 is generated by a manager or a user in advance in a preprocessing step of a mining process, and may extract desired information for each desired time period using the time stamp. In FIG. 4, one day is defined as T1 to T6 by maintaining a time stamp interval of 4 hours each day. Through this, service information according to the location of the mobile user can be extracted at intervals of 4 hours.

5 illustrates an example of integrated log data according to the present invention, in which a single log data is generated by integrating location information of each mobile user and requested service information for each time zone defined in a time stamp defined in a preprocessing process. Fig. 1 shows an example. As shown in FIG. 5, the integrated log data is composed of a user ID, a time stamp, and a behavior pattern, of which a time stamp uses a predefined time stamp (FIG. 4) in a preprocessing process, and the behavior pattern is located. Include the requested service information according to the information. Specifically, in FIG. 5, <A: 1 B: 2 C: 5 D: 8> information is accumulated in the time zone T1 of the user ID 100, where A represents positional information (l in Equation 1) and a number. 1 represents service information (s in Equation 1). The service information is a value set in advance by an administrator or a user, for example, number 1 may be set as a beauty salon and number 2 as a supermarket. At this time, the requested service information according to each location information has sequenced information, which is focused on the fact that the mobile user cannot handle other tasks at once, so that each behavior pattern is not unique but unique at the same time. Has an order. As such, by using the time stamp defined in the preprocessing process, the defined time zone location information and the requested service information may be integrated to generate one consolidated log data. The problem can be solved by limiting the speed of retrieval and the limitation of fast access due to the individualization of information.

As shown in FIG. 5, the integrated log data accumulated in the integrated database is based on a tuple, and is integrated and stored for each user. However, tuple-based data structures cannot meet the need for faster, more accurate and easier to load structures in ubiquitous computing environments. Therefore, in the present invention, the tuple-based data structure is changed to a compact and easy-to-load tree structure to extract useful behavior patterns, that is, frequent behavior patterns of mobile users. Hereinafter, a process of constructing a tree of location information and service information using the log data of FIG. 5 will be described.

FIG. 6 is an exemplary diagram of a header table according to the present invention. FIG. 5 is a diagram illustrating a process of extracting location information satisfying a minimum support degree Min_Sup or more based on information accumulated in an integrated database as shown in FIG. 5. First, as shown in FIG. 6A, information obtained by extracting and counting each positional information generated in FIG. 6 regardless of time is generated, and the positional information of each positional information, that is, the count is greater than or equal to a preset threshold value, is generated. Bay is extracted to generate a header table as shown in FIG. FIG. 6 (b) shows an example in which location information having a support degree of 2 or more is extracted with a threshold value of 2. FIG. In this case, it is judged that F and G do not have frequent support of 1. This is because only the location information above the threshold value is extracted because the service information is not frequent in the frequent location information. At this time, the scan of the integrated database is performed only once, and then the scan of the integrated database is not performed. Thus, the time required to scan the integrated database can be effectively reduced.

Next, a tree is constructed based on the header table generated based on frequent location information as shown in FIG. 6 (b).

7 is a diagram illustrating a tree of location information and service information according to the present invention. In FIG. 7, the tree shown above is a tree constructed based on location information, and the tree shown below is a tree constructed based on service information.

First, a process of generating a location information tree in FIG. 7 will be described. In the location information tree, A, B, C, D, and E are location information included in the consolidated log data of FIG. 5, and the number is a support count of each location information, that is, the number of location information generated as a pattern of the location information tree. Indicates.

1) First, create a root as a null value.

2) The behavior pattern <A: 1 B: 2 C: 5 D: 8> in T1 of the user ID 100 is configured as a tree among the log data of FIG. 5. At this time, A-> B-> C-> D has a sequence as sequence information. This is because mobile users cannot simultaneously do the same thing at the same time.

3) A tree is constructed in the same manner with respect to the location information (A-> D-> E) in T2 of the user ID 100. At this time, since A is already created in the tree, the count is increased to A of the tree generated in 2), and D is newly generated because the information is not in the tree.

4) After the tree is constructed up to T6 for the user ID 100 in the same manner as described above, the tree is constructed in the same manner for the remaining user IDs 101 and 102. In this case, when the location information removed from the header table of FIG. 6 by the support degree is included, such as the location information C-> F in T4 of the user ID 100, the location information C is location information included in the header table. In the count up, the location information F is not included in the header table, so do not create a node.

When the tree for the location information is completed as described above, the tree for the service information is generated and connected to the tail node of the location information tree by a link. FIG. 7 shows only a service information tree for A-> B-> C-> D, A-> B-> E, A-> D-> E of a tree for service information. It was. As shown in Fig. 7, the header table of the service information records a time stamp of a path of the tree. A process of generating a tree for service information is described below.

First, the service tree for D: 3 is constructed from A: 9-> B: 7-> C: 3-> D: 3 of the location information tree. As you can see from D: 3, the count is 3, which means that three service trees are normally needed.

First, a route is generated as a null value, and service information is configured as a tree with respect to the behavior pattern having the location information pattern of A-> B-> C-> D among the integrated log data of FIG. 5. In FIG. 5, the behavior pattern that satisfies these conditions is User ID 100 T1 <A: 1 B: 2 C: 5 D8>, User ID 101 T1 <A: 1 B: 3 C: 5 D: 8>, and User. ID 102 T1 <A: 3 B: 6 C: 8 D: 7>.

First, a tree of service information 1-> 2-> 5-> 8 is formed for user ID 100 T1 <A: 1 B: 2 C: 5 D: 8> and the count is recorded. In Tree 1: 2-> 2: 1-> 5: 1-> 8: 1, the first number represents service information and the second number represents count information. At this time, the level of each node and the time stamp of the node are recorded in the header table. Associate each time stamp in the header table of service information with each node in the service information tree.

Next, a tree is constructed for user ID 101 T1 <A: 1 B: 3 C: 5 D: 8>. In this case, since the service information 1 is a node that has already been created, only the count of the service information is increased, and the next location information 3-> 5-> 8 is a node that has not been created. Create Connect each node of the tree created above and each node of the newly created tree.

Thereafter, a tree for the user ID 102 T1 <A: 3 B: 6 C: 8 D: 7> is constructed. At this time, since service information 3-> 6-> 8-> 7 is a node that has not been created yet, a new tree connected to the root is created. Connect each node of the newly created tree with each node of the previously created tree.

In this way, a tree of service information for the remaining tail nodes is generated.

When a tree for location information and service information is generated as shown in FIG. 7, frequent paths of the generated tree must be extracted to extract useful behavior patterns of mobile users over time.

8 and 9 are diagrams illustrating the extraction of frequent tree paths from the tree of FIG. 7 and an example of extraction of tree paths having a frequency of two or more. As shown in Figs. 8 and 9, only a tree path having a frequency of position information and service information more than a preset value is extracted from the tree of Fig. 9 to extract a frequent tree path.

First, FIG. 8 is a diagram illustrating only a tree path having a frequency of two or more of location information in the tree of FIG. 7. As shown in FIG. 8, in the tree paths E-> D, C-> B, B-> C-> D of the tree of FIG. 7, the tail nodes D, B, and D are frequent views of location information. Is 1 so it is pruned.

Next, FIG. 9 is a diagram of extracting only a tree pass whose frequency of service information is 2 or more in the tree of FIG. 8, wherein the tree path 3 of the service information 3-> 6-> 8-> 7 in time stamp T1. If the service information is 7, the frequency is 1, so it is pruned. Similarly, frequent occurrences when the service information is 2 in the tree paths 1-> 1-> 2 of the service information in the time stamps T2 and T6 are also pruned.

After removing all non-frequent tree paths, a pattern having a minimum support degree or more, as shown in FIG. 9, can be extracted. Through the extracted tree path as shown in FIG. 9, useful behavior patterns frequently occurring at time stamps T1 (06:00 to 10:00) are A: 1, B: 2 (or B: 3), C: 5 and D: You can see that it is 8, and useful patterns of frequent behavior in the time stamps T3 (14:00 ~ 18:00) and T6 (02:00 ~ 06:00) are A: 2, B: 2, and E: 1. have.

In this way, you can find all the useful patterns over time.

While the preferred embodiments of the present invention have been described using specific terms, such descriptions are for illustrative purposes only and it should be understood that various changes and modifications can be made without departing from the spirit and scope of the following claims. do.

1 is a diagram illustrating a system configuration in which an intelligent middleware platform and a mining module are integrated.

2 is a system configuration for extracting a frequent behavior pattern of a mobile user over time according to the present invention.

3 is a flowchart of a process of extracting a frequent behavior pattern of a mobile user over time according to the present invention.

4 is an illustration of a time stamp in accordance with the present invention.

5 is an exemplary diagram of consolidated log data in accordance with the present invention.

6 is an exemplary diagram of a header table according to the present invention.

7 is a diagram illustrating a tree of location information and service information according to the present invention.

8 and 9 are diagrams showing the frequent tree paths extracted from the tree of FIG.

Claims (6)

As a method of extracting a frequent behavior pattern of a mobile user according to a time of converting a behavior pattern of a mobile user's time zone into a tree structure and extracting a frequent tree path, A first step of defining a time stamp for dividing the day into a predetermined time unit; The location information and service information of each mobile user are integrated for each time zone defined in the time stamp, and the mobile user includes a user ID of the mobile user, the time stamp, and service information according to the location information and the location information. A second step of generating integrated log data including a time-phased behavioral pattern for and storing in the integrated database; A third step of scanning the unified database and extracting location information having a support level greater than or equal to a threshold value among behavior patterns of the unified log data to generate a header table; A fourth step of constructing a location information tree for the location information pattern included in the behavior pattern of the integrated log data based on the header table; A fifth step of constructing a service information tree for a service information pattern according to each location information pattern; And And a sixth step of extracting a frequent tree path among the location information tree and the service information tree generated in the fourth and fifth steps. The method of claim 1, The third step, Step 3-1 of scanning the unified database and counting each location information included in the behavior pattern of the unified log data irrespective of the order; And And a step 3-2 of eliminating the location information whose counted information is smaller than the threshold value. The method of claim 1, The fourth step, Step 4-1 of assigning a null value to the root; And A step 4-2 of generating sequenced position information included in the behavior pattern of each time stamp for each user ID of the integrated log data to each node and connecting them to each other; Each node includes a count of occurrences of location information generated in the node and a location information pattern corresponding to a tree path to each node. The method of claim 3, wherein In step 4-2, When the sequenced location information included in the behavior pattern of each time stamp is already generated in a node of the location information tree, the count of the existing node is increased instead of generating a new node. Frequent behavior pattern extraction method. The method of claim 3, wherein The fifth step, Step 5-1 of assigning a null value to the root; Step 5-2 of forming a node with service information of location information patterns included in each tree path of the location information tree among the service information included in the behavior pattern of the integrated log data, and connecting each other; A fifth step of connecting the root of each service information tree to a tail node of a tree path of a corresponding location information tree; And A fifth step of connecting each node with each time stamp of a service information tree header table including a level and a time stamp of each node, And each node of the service tree includes a count of occurrences of service information patterns corresponding to the service information and the tree paths to the respective nodes. The method of claim 5, wherein The sixth step, A step 6-1 of pruning the corresponding tree path and the connected service tree when the count of each node of the tree path is smaller than a predetermined frequency with respect to the location information tree; And And a step 6-2 of pruning the tree path when the count of each node of the tree path is smaller than the frequency of the service information tree. Pattern extraction method.

Images