KR20130026099A - Method and apparatus for creating classifier for spam messages in social networking websites using sender-receiver relationship and method for filtering spam messages - Google Patents

Abstract

PURPOSE: A method for generating a spam message classifier in a social network site by using a relation between a transmitter and a receiver, a device thereof, and a spam message filtering method thereof are provided to accurately filter the spam message by checking if a message is a spam or not through relation information between the receiver and transmitter of the spam message. CONSTITUTION: A spam message classifier generating device collects a real message transceived between members from a specific Internet social network site(S210). The device checks whether the collected message is a spam message(S220). The device calculates a connection degree and a relation distance of a transmitter and a receiver of the message(S230). The device generates a spam message classifier by using features derived from the calculated message(S240). [Reference numerals] (AA) Start; (BB) End; (S210) Collecting messages; (S220) Classifying messages(spam or normal); (S230) Calculating a relation distance and a connection degree between a message receiver and transmitter(using a related member list); (S240) Deriving relation distance and connection degree characteristics of spam messages and generating a spam message classifier by using the same

Description

METHODS AND APPARATUS FOR CREATING CLASSIFIER FOR SPAM MESSAGES IN SOCIAL NETWORKING WEBSITES USING SENDER-RECEIVER RELATIONSHIP AND METHOD FOR FILTERING SPAM MESSAGES }

The present invention relates to a method and apparatus for filtering spam messages in a social network site, and more particularly, to a method and apparatus for filtering spam messages using a relationship between a message sender and a receiver for a message transmitted and received on a social network site. It is about.

Spam is an advertising letter or message sent to an unspecified number of people through e-mail, bulletin boards, text messages, telephones, and the messages feature of Internet portal sites. Such spam is spread indiscriminately, causing inconvenience to spam recipients.

In recent years, the use of Internet sites that provide social network services to strengthen social connections with acquaintances such as friends, friends and friends on the web, build new ones, and form a wider human network has become commonplace. As time goes by, spam messages coming through the site become a problem.

As a conventional method for preventing such spam, a method of classifying a pre-registered specific string or a specific sender account as a spam message is applied. For example, in order to determine spam in an e-mail, the contents of the e-mail are searched to determine whether there is a character string that is determined to be spam. However, in the case of a message received on a social network site having a relatively short text length, there is a problem that it is not easy to determine whether or not the text is spam.

An object of the present invention to solve the above problems is to provide a method for generating a spam message classifier for classifying spam messages from messages received on social network sites.

Another object of the present invention for solving the above problems is to provide an apparatus for generating a spam message classifier for classifying spam messages from messages received on social network sites.

Another object of the present invention to solve the above problems is to provide a method for filtering spam messages from messages received on social network sites.

In order to achieve the above object, the present invention provides a method for generating a spam message classifier using messages collected from a specific Internet social network site and classified as spam messages, by referring to a list of related members stored in a member relations database. For each classified message, calculating a relationship distance based on the number of related members passing through when tracking the sender from the message receiver, and at least possible occurrences of connecting related members passing through from the message receiver to the sender; Calculating a connectivity based on one path, deriving a relation distance and a connectivity characteristic of a spam message using the calculated distance and connectivity between the receiver and the sender, and deriving the derived characteristic. To classify spam messages It provides a spam message classifier generation method comprising the step of generating a classifier for.

Here, the relationship distance and the connection diagram between the receiver and the sender are at least one of the smallest number of related members shown on the path, among possible paths connecting all related members via each of them to reach the sender. It is characterized in that determined based on the shortest path of.

Here, the connection diagram between the receiver and the sender is, when calculating the number of the shortest paths, when the same relationship member appears in two or more shortest paths, only one of the two or more shortest paths is the number of the shortest paths It is characterized by reflecting on.

In this case, the connection degree between the receiver and the sender is calculated using a MIN-CUT method.

Here, when calculating the degree of association between the receiver and the sender, using the RANDOM-WALK algorithm to calculate the weight for the relationship member appearing on each of the shortest path, and to calculate the association with reference to the weight, The weight for the related member may be calculated based on the number of related members included in the related member list of the related member.

Here, the spam message classifier may be generated using any one algorithm among Bagging, LibSVM, FT, J48, and BAyesNet.

In accordance with another aspect of the present invention, a method for filtering a spam message by determining whether a message is a spam message for a specific message received from a specific Internet social network site, and receiving the received message by referring to a list of related members stored in a member relationship database. Calculating a relationship distance based on the number of related members passing through the message receiver when the sender is tracked from the message receiver, and at least one possible connection connecting the related members passing through the message receiver to the sender; Calculating a connectivity based on a route and filtering a message classified as a spam message by applying a spam message classifier to a message for which the relationship distance and connectivity are calculated, wherein the spam message classifier includes Specific internet social net Provides a spam message filtering, wherein design was classified trained using the actual message size within the site.

Here, the relationship distance and the connection diagram between the receiver and the sender are at least one of the smallest number of related members shown on the path, among possible paths connecting all related members via each of them to reach the sender. It is characterized in that determined based on the shortest path of.

Here, the connection diagram between the receiver and the sender is, when calculating the number of the shortest paths, when the same relationship member appears in two or more shortest paths, only one of the two or more shortest paths is the number of the shortest paths It is characterized by reflecting on.

In this case, the connection degree between the receiver and the sender is calculated using a MIN-CUT method.

Here, when calculating the degree of association between the receiver and the sender, by using a RANDOM-WALK algorithm to calculate the weight for the relationship member appearing on each of the shortest path, by calculating the association with reference to the weight, The weight for the related member may be calculated based on the number of related members included in the related member list of the related member.

In accordance with another aspect of the present invention, there is provided a device for generating a spam message classifier using messages collected from a specific Internet social network site and classified as spam messages, wherein the messages classified as spam messages are stored. For each message in the message database with reference to a message database, a member relationship database storing members of the social network site including the recipient of the message and a related member list of the member, and a related member list of the member relationship database. At least one possible linkage unit that calculates a relationship distance based on the number of related members passing through when the sender is traced from the message receiver, and at least one possible linking the related members passing through the message receiver to the sender. Deriving the relationship distance and connectivity characteristics of the spam message using a connectivity calculation unit that calculates a connectivity based on a route, and messages in which the relationship distance and connectivity between the receiver and the sender are calculated, and the derived Provided is a spam message classifier generating device comprising a classifier generator for generating a classifier for classifying spam messages using the property.

Here, the relationship distance and the connection diagram between the receiver and the sender are at least one of the smallest number of related members shown on the path, among possible paths connecting all related members via each of them to reach the sender. It is characterized in that determined based on the shortest path of.

Here, the connection diagram between the receiver and the sender, when calculating the number of the shortest paths, when the same relationship member appears in two or more shortest paths, only one of the two or more shortest paths to the number of the shortest paths It is characterized by calculating using the reflected minimum cut (MIN-CUT) method.

Here, when calculating the degree of association between the receiver and the sender, by using a RANDOM-WALK algorithm to calculate the weight for the relationship member appearing on each of the shortest path, by calculating the association with reference to the weight, The weight for the related member may be calculated based on the number of related members included in the related member list of the related member.

Here, the spam message classifier may be generated using any one algorithm among Bagging, LibSVM, FT, J48, and BAyesNet.

When using a method and apparatus for generating a spam message classifier and a spam message filtering method in a social network site using the relationship between the sender and the receiver according to the present invention, the receiver and the receiver are used by using the relationship information between the message sender and the receiver. The farther the sender's relationship is, the more likely it is that the message is a spam message, thereby filtering spam messages more accurately. In addition, the relationship between the receiver and the sender set in the social network site has an advantage of effectively blocking the spam message sender because it is difficult to operate even the spam message sender.

1 is a conceptual diagram illustrating a schematic configuration of a process of generating a classifier for classifying spam messages according to the present invention.
2 is a sequence chart showing the steps for generating a spam message classifier in accordance with one embodiment of the present invention.
3 is a conceptual diagram illustrating an example of a relationship path between a message receiver and a sender according to an embodiment of the present invention.
4 is a block diagram illustrating a configuration of an apparatus for generating a spam message classifier according to an embodiment of the present invention.
5 is a sequence chart showing a process for filtering a spam message according to an embodiment of the present invention.

While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail. It should be understood, however, that the invention is not intended to be limited to the particular embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention. Like reference numerals are used for like elements in describing each drawing.

The terms first, second, A, B, etc. may be used to describe various elements, but the elements should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component. And / or < / RTI > includes any combination of a plurality of related listed items or any of a plurality of related listed items.

When a component is referred to as being "connected" or "connected" to another component, it may be directly connected to or connected to that other component, but it may be understood that other components may be present in between. Should be. On the other hand, when an element is referred to as being "directly connected" or "directly connected" to another element, it should be understood that there are no other elements in between.

The terminology used herein is for the purpose of describing particular example embodiments only and is not intended to be limiting of the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this application, the terms "comprise" or "have" are intended to indicate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, and one or more other features. It is to be understood that the present invention does not exclude the possibility of the presence or the addition of numbers, steps, operations, components, components, or a combination thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in commonly used dictionaries are to be interpreted as having a meaning consistent with the contextual meaning of the related art and are to be interpreted as either ideal or overly formal in the sense of the present application Do not.

Disclosed is a method for determining whether a message is a spam message based on a relationship between a message sender and a receiver for a message transmitted from a social network site according to the present invention.

The present invention uses the relationship distance and the connection diagram between the sender and the receiver to grasp the relationship between the sender and the receiver. Relationship distances are related to the number of other members of the site that need to pass through to track the message sender from the message recipient. Connection diagrams also relate to the number of instances of members that can be routed in the process of tracking the caller.

In general, the greater the relationship between the message receiver and the sender, that is, the message is more likely to be a spam message when the sender is tracked through several members, rather than when the sender has a direct relationship with the destination. In addition, rather than being able to track the sender through various paths, if the path to track the sender is limited, that is, the smaller the connection, the more likely it is to be a spam message.

Therefore, the present invention is to propose a method of more accurately determining spam messages by analyzing the actual spam messages and deriving the relationship distance and connectivity characteristics between the sender and the receiver.

In the present invention, the min-cut and random-walk methods are used to derive the connectivity characteristics between the sender and the receiver.

The minimum cutting method measures the number of paths between nodes in order to measure the connection degree between two nodes (for example, two members), and the larger the number of paths, the higher the connection degree between two nodes. Meanwhile, when measuring the number of paths between two nodes, there may be an edge that is counted as overlap. In this case, paths without overlapping edges are called edge-independent paths. In the present invention, an edge independent path is measured to measure a more accurate connection degree.

The random walk method is used for PageRank, the algorithm used by search engines. The more the in-link at any node in the graph, the more the inflow link from the node with more inflow links, the higher the score.

Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the accompanying drawings.

1 is a conceptual diagram illustrating a schematic configuration of a process of generating a classifier for classifying spam messages according to the present invention.

Referring to FIG. 1, a schematic configuration for generating a classifier for classifying spam messages according to the present invention includes a data collection step 110, a data classification step 120, and a data training step 130. have.

Referring to FIG. 1, a schematic configuration for generating a classifier for classifying spam messages of the present invention may be described as follows.

The data collection step 110 is a step of collecting actual messages existing in an internet social network site. It is good practice to identify data characteristics that collect as much data as possible, and increase the accuracy of the results. Collected messages also need information about the sender, the recipients, and the relationships with those associated with them, for example, friend relationships. This relationship may be slightly different depending on the policy of the social network site.

For example, in the case of Facebook, the relationship is established through requesting and allowing the relationship between members, and in the case of Twitter, a follower (referred to as another member who requested a relationship with a member) and the member establishes a relationship with another member. If requested, this is a following relationship that refers to the other members above. The present invention is not limited to the process of making a relationship and can be applied to various relationships according to the policy of a social network site.

The data classification step 120 is a step of classifying whether the collected data is spam or normal. This step classifies whether or not spam messages are collected manually. The classified messages are stored in the message database 10 along with their values by calculating the relationship distance and connectivity between the receiver and the sender. That is, each message stored in the message database 10 is stored together with whether it is a spam message or a normal message 11, the relationship distance 12 and the connection diagram 13 between the receiver and the sender.

The data training step 130 is a step of generating a classifier based on the data collected and classified in the previous steps 110 and 120. In this step, we can use the algorithms that exist in the existing field of data mining. For example, Bagging, LibSVM, FT, J48, BayesNet, etc. may be used. Each algorithm uses messages stored in a database to determine the characteristics of the messages in a manner unique to each algorithm, thereby providing an arbitrary message (20). ), A classifier 30 is determined to determine whether the message is a spam message 21 or a normal message 22.

Hereinafter, embodiments of the present invention will be described in more detail with reference to the accompanying drawings. First, a method and apparatus for generating a spam message classifier according to an embodiment of the present invention will be described, and a method for determining whether or not a spam message using the spam message classifier of the present invention will be described.

2 is a sequence chart showing the steps for generating a spam message classifier in accordance with one embodiment of the present invention.

Referring to FIG. 2, the steps for generating a spam message classifier in accordance with an embodiment of the present invention include message collection step S210, message classification step S220, a relationship distance between the receiver and the sender, and a connection degree calculation step S230. And spam message classifier generation step (S240).

In addition, referring to Figure 2, each step of the process for generating a spam message classifier according to an embodiment of the present invention can be described as follows.

Message collecting step (S210) is a step of collecting the actual messages transmitted and received between members from a particular Internet social network site.

The message classification step S220 is a step of determining and classifying whether the messages collected in the message collection step S210 are spam messages. Since this step prepares sample data for training the collected data, it can be done manually for accurate classification.

In the step S230 of calculating the relationship distance and the connection degree between the receiver and the sender, the relationship distance and the connection degree between the receiver and the sender of the corresponding message are calculated for the message classified as a spam message in the above step.

For example, a relationship distance is a measure of whether there is a direct connection between a sender and a receiver or how many people are connected.

At this time, the relationship between the recipient and the sender for each message classified as spam is referred to by referring to the member relationship database in which the members of the social network site and the related members are connected according to the policy of the site. Calculate the distance. A related member may be a friend of the member, a follower, or a cousin. That is, it means other members in the site that are connected to the member through various paths according to the policy of the site, and have a connection relationship such as sending and receiving messages online or viewing member related information.

On the other hand, the relationship between the recipient and the sender of the message may be determined according to the number of related members to reach the sender by backtracking the message sender from the message receiver with reference to the list of related members in the member relations database. have. In addition, the degree of connection between the receiver and the sender may be determined according to the number of possible paths connecting the related members passing through from the receiver to the sender. For example, the greater the number of related members passing through, the greater the relationship distance, and the greater the number of possible paths, the greater the connectivity.

On the other hand, since there may be several related members linked to one member, the tracking path may also vary. Therefore, it is desirable to determine the relationship distance and the connection degree based on the shortest path among the possible paths connecting all the related members who have passed through to the sender with the smallest number of related members shown on the path.

In addition, when calculating the number of shortest paths, when the same related member appears in two or more shortest paths, only one of the two or more shortest paths may be reflected in the number of shortest paths. For this purpose, the connectivity can be calculated using the MIN-CUT method.

As another method of calculating the degree of association between the receiver and the sender, a weight for each member appearing on each shortest path may be calculated using the RANDOM-WALK algorithm, and the degree of association may be calculated with reference to the weight. In this case, the weight for the related member may be calculated based on the number of related members included in the related member list of the related member. In other words, if the number of related members of the member who passes through the shortest path is large, the weight may be set in such a manner that the weight is increased so that the association degree is high and the weight is low.

Alternatively, the above minimum truncation method and the RANDOM-WALK algorithm can be used in combination.

A more detailed method of calculating the relationship distance and the connection between the receiver and the sender will be described later.

Spam message classifier generation step (S250) derives the relation distance and connectivity characteristics of the spam message by using the calculated relation distance and connectivity between the receiver and the sender, and classifies the spam message using the derived characteristics In this step, a classifier is generated.

At this time, the spam message classifier may be generated using any one of algorithms such as Bagging, LibSVM, FT, J48, and BAyesNet.

Hereinafter, a method of generating a relationship distance and a connection diagram between a message sender and a receiver will be described in more detail.

3 is a conceptual diagram illustrating an example of a relationship path between a message receiver and a sender according to an embodiment of the present invention.

(A) of FIG. 3 is a graph in which the number of edge-independent paths having a relation distance of 3 is 3, and (b) shows a graph in which the edge-independent path is 1 having a relation distance of 4.

For example, suppose that a member named F sends a message to a member named A, as shown in FIG.

 To determine the relationship distance between receivers A and F, determine whether A and F are directly related and how many members can be traced to F. For example, determine whether you are a friend, a friend of a friend, or a complete person.

At this time, a method of identifying such a relationship is to determine whether F exists in the related member list of A (hereinafter, referred to as friend list). That is, if F is present in the friend list of A in FIG. 3 (a), the relationship distance may be 1. However, if F does not exist in A's friend list, all friends in A's friend list are searched for each friend's friend list to see if F exists. If F is in the friend list of A's friends (B, C, D, and E), the relationship distance can be 2.

However, if F is not present in A's friend's friend list, the process of tracking F is repeated for all friends of A's friend list. On the other hand, the sender tracking process can put unnecessary load on the system when the sender cannot track or if there are too many friends. Therefore, the maximum tracking level is specified. You can also specify the minimum value (for example, 5) to set the relationship distance.

3 (a) is a graph showing a result of tracing F existing in a friend list of B, C, D, and E which are friends of A, although F does not exist in A's friend list.

At this time, in order to track F from A, two edges are connected for each path through one of the members B, C, D, and E in the middle of the shortest path of A and F as shown in the graph 301 (311-). 312, 321-322, 331-332, 341-342), so the relationship between A and F is 2.

On the other hand, there are four edge-independent shortest paths between A and F: ABF (311-312), ACF (321-322), ADF (331-332), and AEF (341-342), so the connectivity can be four. have.

As another example, referring to FIG. 3B, it is assumed that a Q member sent a message to a H member, and traces Q back from the message receiver H.

At this time, in order to track Q from H, six shortest paths of H and Q (H-I-M-P-Q, H-J-M-P-Q, H-J-N-P-Q, H-K-N-P-Q, H-K-O-P-Q, and H-L-O-P-Q) may appear. Since four edges are connected for each of these shortest paths, the distance between H and Q is four.

In addition, since there are overlapping edges 353, 364, 383, and 354 in these shortest paths, there is only one edge-independent path except for paths having overlapping edges, so that H and Q have a connection degree of 1.

4 is a block diagram illustrating a configuration of an apparatus for generating a spam message classifier according to an embodiment of the present invention.

Referring to FIG. 4, an apparatus for generating a spam message classifier in accordance with an embodiment of the present invention includes a message collector 410, a message classifier 420, a relationship distance and connectivity calculator 430, and a spam message classification. The child generator 440 includes a message database 10 and a membership relationship database 90.

In addition, referring to Figure 4 the apparatus for generating a spam message classifier according to an embodiment of the present invention can be described as follows.

The message collecting unit 410 collects actual messages transmitted and received between members from a specific internet social network site.

The message classifier 420 is a part of determining and classifying whether or not the messages collected by the message collector 410 are spam messages and storing them in the message database 10.

The relationship distance and connectivity calculation unit 430 calculates a relationship distance and a connection between the receiver and the sender of the corresponding message for the message classified as spam messages stored in the message database 10.

For example, a relationship distance is a measure of whether there is a direct connection between a sender and a receiver or how many people are connected.

At this time, the recipient and the sender for each message classified as spam by referring to the member relations database 90 in which the members of the corresponding social network site and the related members are connected according to the policy of the site are stored. Calculate the relationship distance between. A related member may be a friend of the member, a follower, or a cousin. That is, it means other members in the site that are connected with the member through various paths according to the policy of the site, and have a connection relationship formed according to the site policy such as sending and receiving messages online or viewing member related information.

On the other hand, the relationship distance between the receiver and the sender of the message is based on the number of related members to reach the sender by backtracking the message sender from the message receiver with reference to the related member list in the membership relations database 90. Can be determined accordingly. In addition, the degree of connection between the receiver and the sender may be determined according to the number of possible paths connecting the related members passing through from the receiver to the sender. For example, the greater the number of related members passing through, the greater the relationship distance, and the greater the number of possible paths, the greater the connectivity.

In addition, when calculating the number of shortest paths, when the same related member appears in two or more shortest paths, only one of the two or more shortest paths may be reflected in the number of shortest paths. For this purpose, the connectivity can be calculated using the MIN-CUT method.

As another method of calculating the degree of association between the receiver and the sender, a weight for each member appearing on each shortest path may be calculated using the RANDOM-WALK algorithm, and the degree of association may be calculated with reference to the weight. In this case, the weight for the related member may be calculated based on the number of related members included in the related member list of the related member. In other words, if the number of related members of the transit member on the shortest path is large, the weight may be set in such a manner that the weight is increased so that the association degree is high and the weight is low. Alternatively, the above minimum truncation method and the RANDOM-WALK algorithm can be used in combination.

The spam message classifier generation unit 440 derives the relation distance and connectivity characteristics of the spam message using the calculated relation distance and connectivity between the receiver and the sender, and classifies the spam message using the derived characteristics. This is the part that creates the classifier. At this time, the spam message classifier may be generated using any one of algorithms such as Bagging, LibSVM, FT, J48, and BAyesNet.

Hereinafter, a method of determining whether or not a spam message is for a random message using a spam message classifier according to the present invention will be described.

5 is a sequence chart showing a process for filtering a spam message according to an embodiment of the present invention.

Referring to FIG. 5, each step for filtering a spam message according to an embodiment of the present invention includes a message receiving step (S510), a relationship distance between the receiver and the sender, a connection degree calculation step (S520), and a spam message filtering step (S530). It is configured to include).

Message receiving step (S510) is a step for receiving a message to determine whether a spam message for any message sent from any member in a particular Internet social network site.

The relationship distance between the receiver and the sender and the connection degree calculation step (S520) is a member relationship in which a social network site member and a relationship member list having a connection relationship according to the policy of the site are stored for any message received. Computing the relationship distance and connectivity between the receiver and the sender with respect to the received message by referring to the database.

At this time, the relationship distance between the receiver and the sender is determined according to the number of related members to reach the caller by backtracking the caller from the receiver using the related member list, and the connection degree is reached until reaching the caller. It can be determined by the number of possible paths connecting the related members via.

  Since a more detailed method of calculating the relation distance and the connection diagram has been described above, it will be omitted.

The spam message filtering step S530 is a step of filtering a message determined to be a spam message by applying a spam message classifier to a message whose relationship distance and connectivity are calculated. At this time, the spam message classifier used uses a classifier trained by using the messages of which the spam message according to the present invention and the relation distance and the connection degree between the receiver and the sender have been calculated.

It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit or scope of the present invention as defined by the following claims It can be understood that

Claims (16)

A method for generating a spam message classifier using messages collected from a specific internet social network site and classified as spam messages,
Calculating a relationship distance based on the number of related members passing through the message from the message receiver for each of the classified messages by referring to the related member list stored in the member relations database;
Calculating a connectivity based on at least one possible path connecting the related members from the message receiver to the sender; And
Deriving a relationship distance and a connection property of a spam message using the calculated values of the relationship distance and the connection between the receiver and the sender, and generating a classifier for classifying the spam message using the derived property. Spam message classifier generation method comprising the. The method of claim 1,
The relationship distance and the connection diagram between the receiver and the sender are at least one shortest number of related members represented on the path among the possible paths connecting all related members each passing through the caller. Spam message classifier generation method characterized in that determined based on the path. The method of claim 2,
The connection diagram between the receiver and the sender is,
When calculating the number of shortest paths, if the same related member appears in two or more shortest paths, a spam message classifier is generated, wherein only one of the two or more shortest paths is reflected in the number of shortest paths. Way. The method of claim 3, wherein
The method for generating a spam message classifier, characterized in that the degree of connection between the receiver and the sender is calculated using a MIN-CUT method. 3. The method according to claim 2 or 3,
In calculating the degree of association between the receiver and the sender,
The RANDOM-WALK algorithm is used to calculate weights for the related members appearing on each of the shortest paths, and calculates the association with reference to the weights.
And the weight for the related member is calculated based on the number of related members included in the related member list of the related member. The method of claim 1,
The spam message classifier is a spam message classifier generation method, characterized in that generated using any one of the algorithm of Bagging, LibSVM, FT, J48 and BAyesNet. A method of filtering spam messages by determining whether a message is spam for a specific message received from a specific internet social network site.
Calculating a relationship distance based on the number of related members passing through the received message when tracking a sender from the message receiver with respect to the received message by referring to the related member list stored in the member relations database;
Calculating a connectivity based on at least one possible path connecting the related members from the message receiver to the sender; And
Filtering a message classified as a spam message by applying a spam message classifier to the message whose relationship distance and connectivity are calculated;
And wherein said spam message classifier is a classifier trained using actual messages in said particular internet social network site. 8. The method of claim 7,
The relationship distance and the connection diagram between the receiver and the sender are at least one shortest number of related members represented on the path among the possible paths connecting all related members each passing through the caller. Spam message filtering method characterized in that determined based on the path. The method of claim 8,
The connection diagram between the receiver and the sender is,
When calculating the number of the shortest paths, if the same related member appears in two or more shortest paths, only one of the two or more shortest paths is reflected in the number of shortest paths. The method of claim 9,
The method for filtering spam messages, characterized in that the degree of connection between the receiver and the sender is calculated using a MIN-CUT method. The method according to claim 7 or 8,
In calculating the degree of association between the receiver and the sender,
By using the RANDOM-WALK algorithm to calculate the weight for the relationship member appearing on each of the shortest path, the association degree is calculated with reference to the weight,
And the weight for the related member is calculated based on the number of related members included in the related member list of the related member. A device for generating a spam message classifier using messages collected from a specific Internet social network site and classified as spam messages,
A message database storing a message classified as the spam message;
A member relations database storing members of the social network site including the recipient of the message and a related member list of the member;
A relationship distance calculation unit that calculates a relationship distance based on the number of relationship members passing through each message in the message database by referring to a relationship member list of the member relationship database when the sender is traced from the message receiver;
A connectivity calculation unit configured to calculate a connectivity based on at least one possible path connecting the related members passing through the message receiver to the sender; And
Generating a classifier that derives the relation distance and connectivity characteristics of the spam message by using the messages whose relationship distance and connectivity between the receiver and the sender are calculated, and generates a classifier that classifies the spam messages using the derived characteristics Spam message classifier generator, characterized in that it comprises a wealth. 13. The method of claim 12,
The relationship distance and the connection diagram between the receiver and the sender are at least one shortest number of related members represented on the path among the possible paths connecting all related members each passing through the caller. Spam message classifier generator, characterized in that determined based on the path. The method of claim 13,
The connection diagram between the receiver and the sender reflects only one of the two or more shortest paths to the number of the shortest paths when the same related member appears in two or more shortest paths when calculating the number of the shortest paths. Spam message classifier generator, characterized in that calculated using the MIN-CUT method. The method according to claim 12 or 13,
In calculating the degree of association between the receiver and the sender,
By using the RANDOM-WALK algorithm to calculate the weight for the relationship member appearing on each of the shortest path, the association degree is calculated with reference to the weight,
And a weight for the related member is calculated based on the number of related members included in the related member list of the related member. 13. The method of claim 12,
The spam message classifier generating apparatus for spam message classifier, characterized in that generated using any one of the algorithm of Bagging, LibSVM, FT, J48 and BAyesNet.

Images